Upload folder using huggingface_hub

README.md

@@ -10,23 +10,33 @@ app_file: space.py
 ---
 
 # `gradio_livelog`
-<img alt="Static Badge" src="https://img.shields.io/badge/version%20-%200.0.1%20-%20blue"> <a href="https://huggingface.co/spaces/elismasilva/gradio_livelog"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Demo-blue"></a><p><span>💻 <a href='https://github.com/DEVAIEXP/gradio_component_livelog'>Component GitHub Code</a></span></p>
+<img alt="Static Badge" src="https://img.shields.io/badge/version%20-%200.0.3%20-%20blue"> <a href="https://huggingface.co/spaces/elismasilva/gradio_livelog"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Demo-blue"></a><p><span>💻 <a href='https://github.com/DEVAIEXP/gradio_component_livelog'>Component GitHub Code</a></span></p>
 
 A Live Log Component for Gradio Interface
 
 ## Key Features
 
-`LiveLog` provides a rich, terminal-like experience directly in your Gradio UI.
+`LiveLog` elevates Gradio applications by providing a powerful, terminal-like monitoring experience directly in your UI. It's designed for both simple progress tracking and complex pipeline introspection.
 
-- **Real-time Log Streaming:** Display log messages as they are generated in your Python backend, perfect for monitoring long-running tasks.
-- **Integrated Progress Bar:** Track the progress of loops and processes with a familiar, `tqdm`-style progress bar. 
-    - Includes iterations per second (it/s) calculation.
-    - The bar automatically changes color to green on success or red on error for clear visual feedback.
-- **Automatic `logging` Capture:** Effortlessly capture all logs generated by Python's standard `logging` module (`logging.info`, `logging.error`, etc.) using the `capture_logs` context manager.
-- **Flexible Display Modes:** Choose to display the full component (logs + progress bar), logs only, or the progress bar only to fit different UI layouts.
-- **Highly Customizable:** Tailor the component's appearance with properties for background color, line numbers, and autoscrolling.
-- **Utility Controls:** Comes with built-in buttons to clear the log, copy all content to the clipboard, and download logs as a text file.
-- **Console Silencing:** Optionally suppress log output in your Python console to keep it clean while still displaying everything in the UI.
+-   **Dual-Mode Progress Tracking:** A sophisticated progress bar that operates in two modes for maximum accuracy:
+    -   **Internal Rate Calculation:** For simple loops, it features a built-in, `tqdm`-style progress calculator with **Exponential Moving Average (EMA)** smoothing for a stable and realistic `it/s` or `s/it` display.
+    -   **External `tqdm` Capture:** For deep integration, it can **directly capture and display the *exact* rate** from an existing `tqdm` instance running inside a backend library (like `diffusers`). This eliminates measurement overhead and provides a perfectly synchronized view of your pipeline's true performance.
+
+-   **Rich, Real-time Log Streaming:** Display log messages as they are generated.
+    -   Supports standard Python log levels (`INFO`, `WARNING`, `ERROR`, etc.) with corresponding colors.
+    -   Includes support for **custom log levels** (like "SUCCESS") for enhanced visual feedback.
+
+-   **Advanced Multi-Logger Capture:** The `capture_logs` utility is designed for complex applications.
+    -   Effortlessly capture logs from **multiple, independent Python loggers** simultaneously (e.g., your app's logger and a library's internal logger).
+    -   Correctly handles logger hierarchies and propagation, making it robust for any logging setup.
+
+-   **Flexible Display & Layout Control:** Adapt the component to any UI layout.
+    -   **Three Display Modes:** Show the full component (`logs + progress`), `logs only`, or `progress bar only`.
+    -   Highly customizable appearance with properties for `height`, `background_color`, `line_numbers`, and `autoscrolling`.
+
+-   **Comprehensive Utility Controls:**
+    -   Built-in header buttons to **Clear**, **Copy**, and **Download** log content.
+    -   Optionally suppress log output in your Python console to keep it clean while still displaying everything in the UI.
 
 ## Installation
 
@@ -39,9 +49,9 @@ pip install gradio_livelog
 ```python
 # demo/app.py
 
+import sys
 import gradio as gr
 import torch
-import time
 import logging
 import random
 import numpy as np
@@ -53,24 +63,28 @@ import spaces
 
 # Import the component and ALL its utilities
 from gradio_livelog import LiveLog
-from gradio_livelog.utils import ProgressTracker, capture_logs
+from gradio_livelog.utils import ProgressTracker, Tee, TqdmToQueueWriter, capture_logs
 
 # --- 1. SETUP ---
 MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
 MAX_SEED = np.iinfo(np.int32).max
 
 # --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
-logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
-
-async def run_process(disable_console: bool, run_error_case: bool):
-    with capture_logs(disable_console=disable_console) as get_logs:
+app_logger = logging.getLogger("logging_app")
+app_logger.setLevel(logging.INFO)
+console_handler = logging.StreamHandler()
+console_handler.flush = sys.stderr.flush
+app_logger.addHandler(console_handler)
+
+async def run_process(disable_console: bool, rate_unit: str, run_error_case: bool):
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
         total_steps = 100
-        tracker = ProgressTracker(total=total_steps, description="Simulating a process...")
+        tracker = ProgressTracker(total=total_steps, description="Simulating a process...", rate_unit=rate_unit)
         all_logs = []
         last_log_content = None
         
         initial_log = f"Starting simulated process with {total_steps} steps..."
-        logging.info(initial_log)
+        app_logger.info(initial_log)
         logs = [
             {
                 "type": "log",
@@ -88,14 +102,14 @@ async def run_process(disable_console: bool, run_error_case: bool):
             current_step = i + 1
             
             if current_step == 10:
-                logging.warning(f"Low disk space warning at step {current_step}.")
+                app_logger.warning(f"Low disk space warning at step {current_step}.")
             elif current_step == 30:
-                logging.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
+                app_logger.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
             elif current_step == 75:
-                logging.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
+                app_logger.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
             
             if run_error_case and current_step == 50:
-                logging.error("A fatal simulation error occurred! Aborting.")
+                app_logger.error("A fatal simulation error occurred! Aborting.")
                 logs = [
                     {
                         "type": "log",
@@ -123,7 +137,7 @@ async def run_process(disable_console: bool, run_error_case: bool):
             yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
         
         final_log = "Process completed successfully!"
-        logging.log(logging.INFO + 5, final_log)
+        app_logger.log(logging.INFO + 5, final_log)
         logs = [
             {
                 "type": "log",
@@ -142,47 +156,60 @@ def update_livelog_properties(mode, color, lines, scroll):
 def clear_output():
     return None
 
-async def run_success_case(disable_console: bool):
+async def run_success_case(disable_console: bool, rate_unit: str):
     yield None    
-    async for update in run_process(disable_console=disable_console, run_error_case=False):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=False):
         yield update
 
-async def run_error_case(disable_console: bool):
+async def run_error_case(disable_console: bool, rate_unit: str):
     yield None
-    async for update in run_process(disable_console=disable_console, run_error_case=True):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=True):
         yield update
 
 
 # --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
 diffusion_pipeline = None
+pipeline_lock = threading.Lock()
 def load_pipeline(on_load=True):
     """A function to load the model, ensuring it's only done once."""
     global diffusion_pipeline
-    if diffusion_pipeline is None:
-        print("Loading Stable Diffusion model for the first time...")
-        pipe = StableDiffusionXLPipeline.from_pretrained(
-            MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
-        )
-        pipe.enable_vae_tiling()
-        pipe.enable_model_cpu_offload()
-        pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-        pipe.set_progress_bar_config(disable=True)
-        diffusion_pipeline = pipe
-        print("Model loaded successfully.")
-    
+    with pipeline_lock:
+        if diffusion_pipeline is None:
+            print("Loading Stable Diffusion model for the first time...")
+            pipe = StableDiffusionXLPipeline.from_pretrained(
+                MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
+            )
+            pipe.enable_vae_tiling()
+            pipe.enable_model_cpu_offload()
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)        
+            diffusion_pipeline = pipe
+            print("Model loaded successfully!")
+        
     if not on_load:
         return diffusion_pipeline
 
 @spaces.GPU(duration=60, enable_queue=True)
-def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
+def run_diffusion_in_thread(prompt: str, disable_console: bool, update_queue: queue.Queue):
     """
     This function now uses capture_logs to listen to internal diffusers logs
     while retaining the superior data structure you designed.
     """
     tracker = None    
-    with capture_logs() as get_logs:
-        try:
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        try:            
             pipe = load_pipeline(on_load=False)
+            
+            #We will capture pipeline tqdm s/it progress instead            
+            rate_queue = queue.Queue()
+            tqdm_writer = TqdmToQueueWriter(rate_queue)
+            
+            progress_bar_handler = Tee(sys.stderr, tqdm_writer)
+            pipe.set_progress_bar_config(file=progress_bar_handler,  #if you dont need to see the tqdm progress on console set file=tqdm_writer instead            
+                                        disable=False,  
+                                        ncols=100,
+                                        dynamic_ncols=True,
+                                        ascii=" █")
+            
             seed = random.randint(0, MAX_SEED)
             generator = torch.Generator(device="cuda").manual_seed(seed)
             prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
@@ -190,67 +217,110 @@ def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
             num_inference_steps = 10
             
             all_logs = []
-            last_log_content = None
+            last_known_rate_data = None
 
             # Helper function to process and store new logs
-            def process_and_store_logs():
-                nonlocal all_logs, last_log_content
+            def process_and_send_updates(status="running", advance=0, final_image_payload=None):
+                """
+                This is the core callback function. It captures new logs, formats them,
+                and sends a complete update object (logs + progress) to the UI queue.
+                This should also be called after the log record.
+                """
+                nonlocal all_logs, last_known_rate_data
+                new_rate_data = None
+                while not rate_queue.empty():
+                    try:
+                        new_rate_data = rate_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                
+                if new_rate_data is not None:
+                    last_known_rate_data = new_rate_data
+                
                 new_records = get_logs()
                 if new_records:
-                    new_logs = [{"type": "log", "level": r.levelname, "content": r.getMessage()} for r in new_records]
+                    new_logs = [{
+                        "type": "log",
+                        "level": "SUCCESS" if r.levelno == logging.INFO + 5 else r.levelname,
+                        "content": r.getMessage()
+                    } for r in new_records]
                     all_logs.extend(new_logs)
-                    last_log_content = all_logs[-1]["content"]
-            
-            logging.info(f"Using seed: {seed}")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            logging.info("Starting diffusion process...")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps")
+                
+                # Use the tracker to generate the progress update dictionary if it exists.
+                # If not, create a preliminary update dictionary.
+                update_dict = {}
+                
+                if tracker:
+                    update_dict = tracker.update(
+                        advance=advance, 
+                        status=status, 
+                        logs=all_logs,
+                        rate_data=last_known_rate_data                               
+                    )
+                else:
+                    # Initial state before the tracker is created.
+                    update_dict = {
+                        "type": "progress", 
+                        "logs": all_logs, 
+                        "current": 0, 
+                        "total": num_inference_steps, 
+                        "desc": "Diffusion Steps" # Description is sent once
+                    }
+
+                # Put the update on the queue. The image payload is usually None
+                # until the very end.
+                update_queue.put((final_image_payload, update_dict))
+                
+            app_logger.info(f"Using seed: {seed}")
+            process_and_send_updates()
+                        
+            app_logger.info("Starting diffusion process...")
+            process_and_send_updates()
+                        
+            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps", rate_unit='it/s')
             
             def progress_callback(pipe_instance, step, timestep, callback_kwargs):
-                process_and_store_logs() # Check for new logs from diffusers at each step
-                update_dict = tracker.update(logs=all_logs)
-                update_queue.put((None, update_dict))
+                process_and_send_updates(advance=1) 
                 return callback_kwargs
-            
+                        
             images = pipe(
                 prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
                 guidance_scale=3.0, num_inference_steps=num_inference_steps,
                 generator=generator, callback_on_step_end=progress_callback
             ).images
             
-            logging.log(logging.INFO + 5, "Image generated successfully!")
-            process_and_store_logs()
-            
-            final_update = tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-            update_queue.put((images, final_update))
+            app_logger.log(logging.INFO + 5, "Image generated successfully!")
+            process_and_send_updates(status="success", final_image_payload=images)
+           
 
         except Exception as e:
-            logging.error(f"Error in diffusion thread: {e}", exc_info=True)
-            process_and_store_logs() # Capture the final error log
-            if tracker:
-                error_update = tracker.update(advance=0, status="error", logs=all_logs, log_content=f"An error occurred: {e}")
-                update_queue.put((None, error_update))
+            app_logger.error(f"Error in diffusion thread: {e}, process aborted!", exc_info=True)                    
+            process_and_send_updates(status="error")                                                        
         finally:
             update_queue.put(None)
             
+            
 @spaces.GPU(duration=60, enable_queue=True)
 def generate(prompt):
     """This function starts the worker thread and yields updates from the queue."""   
-    yield None, None
-    yield None, {"type": "log", "level": "INFO", "content": "Preparing generation..."}
+    yield None, None, gr.update(interactive=False)    
     update_queue = queue.Queue()
-    diffusion_thread = threading.Thread(target=run_diffusion_in_thread, args=(prompt, update_queue))
+    diffusion_thread = threading.Thread(target=run_diffusion_in_thread,  args=(prompt, False, update_queue))
     diffusion_thread.start()
+    final_images = None
     while True:
         update = update_queue.get()
-        if update is None: break
-        yield update
-
+        if update is None: 
+            break
+        
+        images, log_update = update
+        
+        if images:
+            final_images = images
+      
+        yield final_images, log_update, gr.skip()
+    
+    yield final_images, log_update, gr.update(interactive=True)
 
 # --- 4. THE COMBINED GRADIO UI with TABS ---
 with gr.Blocks(theme=gr.themes.Ocean()) as demo:
@@ -269,17 +339,18 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                     with gr.Group():
                         gr.Markdown("### Component Properties")
                         display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
+                        rate_unit = gr.Radio(["it/s","s/it"], label="Progress rate unit", value="it/s")
                         bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
                         line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
                         autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
+                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                     with gr.Group():
                         gr.Markdown("### Simulation Controls")
-                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                         start_btn = gr.Button("Run Success Case", variant="primary")
                         error_btn = gr.Button("Run Error Case")
             
-            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox], outputs=feature_logger)
+            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
             feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
             controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
             for control in controls:
@@ -306,7 +377,7 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                         label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
                     )
             
-            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
+            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer, run_button])
             prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
             livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
             

app.py

@@ -1,278 +1,339 @@
-# demo/app.py
-
-import gradio as gr
-import torch
-import time
-import logging
-import random
-import numpy as np
-from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
-import threading
-import queue
-import asyncio
-import spaces
-
-# Import the component and ALL its utilities
-from gradio_livelog import LiveLog
-from gradio_livelog.utils import ProgressTracker, capture_logs
-
-# --- 1. SETUP ---
-MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
-MAX_SEED = np.iinfo(np.int32).max
-
-# --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
-logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
-
-async def run_process(disable_console: bool, run_error_case: bool):
-    with capture_logs(disable_console=disable_console) as get_logs:
-        total_steps = 100
-        tracker = ProgressTracker(total=total_steps, description="Simulating a process...")
-        all_logs = []
-        last_log_content = None
-        
-        initial_log = f"Starting simulated process with {total_steps} steps..."
-        logging.info(initial_log)
-        logs = [
-            {
-                "type": "log",
-                "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
-                "content": record.getMessage()
-            }
-            for record in get_logs()
-        ]
-        all_logs.extend(logs)
-        last_log_content = logs[-1]["content"] if logs else None        
-        yield tracker.update(advance=0, status="running", logs=all_logs, log_content=None)
-
-        for i in range(total_steps):
-            await asyncio.sleep(0.03)
-            current_step = i + 1
-            
-            if current_step == 10:
-                logging.warning(f"Low disk space warning at step {current_step}.")
-            elif current_step == 30:
-                logging.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
-            elif current_step == 75:
-                logging.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
-            
-            if run_error_case and current_step == 50:
-                logging.error("A fatal simulation error occurred! Aborting.")
-                logs = [
-                    {
-                        "type": "log",
-                        "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
-                        "content": record.getMessage()
-                    }
-                    for record in get_logs()
-                ]
-                all_logs.extend(logs)
-                last_log_content = logs[-1]["content"] if logs else last_log_content
-                yield tracker.update(advance=0, status="error", logs=all_logs, log_content=last_log_content)
-                return
-            
-            logs = [
-                {
-                    "type": "log",
-                    "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
-                    "content": record.getMessage()
-                }
-                for record in get_logs()
-            ]
-            all_logs.extend(logs)
-            if logs:
-                last_log_content = logs[-1]["content"]
-            yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
-        
-        final_log = "Process completed successfully!"
-        logging.log(logging.INFO + 5, final_log)
-        logs = [
-            {
-                "type": "log",
-                "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
-                "content": record.getMessage()
-            }
-            for record in get_logs()
-        ]
-        all_logs.extend(logs)
-        last_log_content = logs[-1]["content"] if logs else last_log_content
-        yield tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-        
-def update_livelog_properties(mode, color, lines, scroll):
-    return gr.update(display_mode=mode, background_color=color, line_numbers=lines, autoscroll=scroll)
-
-def clear_output():
-    return None
-
-async def run_success_case(disable_console: bool):
-    yield None    
-    async for update in run_process(disable_console=disable_console, run_error_case=False):
-        yield update
-
-async def run_error_case(disable_console: bool):
-    yield None
-    async for update in run_process(disable_console=disable_console, run_error_case=True):
-        yield update
-
-
-# --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
-diffusion_pipeline = None
-def load_pipeline(on_load=True):
-    """A function to load the model, ensuring it's only done once."""
-    global diffusion_pipeline
-    if diffusion_pipeline is None:
-        print("Loading Stable Diffusion model for the first time...")
-        pipe = StableDiffusionXLPipeline.from_pretrained(
-            MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
-        )
-        pipe.enable_vae_tiling()
-        pipe.enable_model_cpu_offload()
-        pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-        pipe.set_progress_bar_config(disable=True)
-        diffusion_pipeline = pipe
-        print("Model loaded successfully.")
-    
-    if not on_load:
-        return diffusion_pipeline
-
-@spaces.GPU(duration=60, enable_queue=True)
-def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
-    """
-    This function now uses capture_logs to listen to internal diffusers logs
-    while retaining the superior data structure you designed.
-    """
-    tracker = None    
-    with capture_logs() as get_logs:
-        try:
-            pipe = load_pipeline(on_load=False)            
-            seed = random.randint(0, MAX_SEED)
-            generator = torch.Generator(device="cuda").manual_seed(seed)
-            prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
-            negative_prompt_style = "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly"
-            num_inference_steps = 10
-            
-            all_logs = []
-            last_log_content = None
-
-            # Helper function to process and store new logs
-            def process_and_store_logs():
-                nonlocal all_logs, last_log_content
-                new_records = get_logs()
-                if new_records:
-                    new_logs = [{"type": "log", "level": r.levelname, "content": r.getMessage()} for r in new_records]
-                    all_logs.extend(new_logs)
-                    last_log_content = all_logs[-1]["content"]
-            
-            logging.info(f"Using seed: {seed}")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            logging.info("Starting diffusion process...")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps")
-            
-            def progress_callback(pipe_instance, step, timestep, callback_kwargs):
-                process_and_store_logs() # Check for new logs from diffusers at each step
-                update_dict = tracker.update(logs=all_logs)
-                update_queue.put((None, update_dict))
-                return callback_kwargs
-            
-            images = pipe(
-                prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
-                guidance_scale=3.0, num_inference_steps=num_inference_steps,
-                generator=generator, callback_on_step_end=progress_callback
-            ).images
-            
-            logging.log(logging.INFO + 5, "Image generated successfully!")
-            process_and_store_logs()
-            
-            final_update = tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-            update_queue.put((images, final_update))
-
-        except Exception as e:
-            logging.error(f"Error in diffusion thread: {e}", exc_info=True)
-            process_and_store_logs() # Capture the final error log
-            if tracker:
-                error_update = tracker.update(advance=0, status="error", logs=all_logs, log_content=f"An error occurred: {e}")
-                update_queue.put((None, error_update))
-        finally:
-            update_queue.put(None)
-            
-@spaces.GPU(duration=60, enable_queue=True)
-def generate(prompt):
-    """This function starts the worker thread and yields updates from the queue."""   
-    yield None, None
-    yield None, {"type": "log", "level": "INFO", "content": "Preparing generation..."}
-    update_queue = queue.Queue()
-    diffusion_thread = threading.Thread(target=run_diffusion_in_thread, args=(prompt, update_queue))
-    diffusion_thread.start()
-    while True:
-        update = update_queue.get()
-        if update is None: break
-        yield update
-
-
-# --- 4. THE COMBINED GRADIO UI with TABS ---
-with gr.Blocks(theme=gr.themes.Ocean()) as demo:
-    gr.HTML("<h1><center>LiveLog Component Showcase</center></h1>")
-
-    with gr.Tabs():
-        with gr.TabItem("LiveLog Feature Demo"):            
-            gr.Markdown("### Test all features of the LiveLog component interactively.")
-            with gr.Row():
-                with gr.Column(scale=3):
-                    feature_logger = LiveLog(
-                        label="Process Output", line_numbers=True, height=550,
-                        background_color="#000000", display_mode="full"
-                    )
-                with gr.Column(scale=1):
-                    with gr.Group():
-                        gr.Markdown("### Component Properties")
-                        display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
-                        bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
-                        line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
-                        autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
-                    with gr.Group():
-                        gr.Markdown("### Simulation Controls")
-                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
-                        start_btn = gr.Button("Run Success Case", variant="primary")
-                        error_btn = gr.Button("Run Error Case")
-            
-            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
-            controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
-            for control in controls:
-                control.change(fn=update_livelog_properties, inputs=controls, outputs=feature_logger)
-        
-        with gr.TabItem("Diffusion Pipeline Integration"):               
-            gr.Markdown("### Use `LiveLog` to monitor a real image generation process.")
-            with gr.Row():
-                with gr.Column(scale=3):
-                    with gr.Group():
-                        prompt = gr.Textbox(
-                            label="Enter your prompt", show_label=False,
-                            placeholder="A cinematic photo of a robot in a floral garden...",
-                            scale=8, container=False
-                        )
-                        run_button = gr.Button("Generate", scale=1, variant="primary")
-                    
-                    livelog_viewer = LiveLog(
-                        label="Process Monitor", height=250, display_mode="full", line_numbers=False
-                    )
-                
-                with gr.Column(scale=2):
-                    result_gallery = gr.Gallery(
-                        label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
-                    )
-            
-            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
-            prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
-            livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
-            
-    # This ensures the model is downloaded/loaded once when the app starts.
-    demo.load(load_pipeline, None, None)
-                
-if __name__ == "__main__":
+# demo/app.py
+
+import sys
+import gradio as gr
+import torch
+import logging
+import random
+import numpy as np
+from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
+import threading
+import queue
+import asyncio
+import spaces
+
+# Import the component and ALL its utilities
+from gradio_livelog import LiveLog
+from gradio_livelog.utils import ProgressTracker, Tee, TqdmToQueueWriter, capture_logs
+
+# --- 1. SETUP ---
+MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
+MAX_SEED = np.iinfo(np.int32).max
+
+# --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
+app_logger = logging.getLogger("logging_app")
+app_logger.setLevel(logging.INFO)
+console_handler = logging.StreamHandler()
+console_handler.flush = sys.stderr.flush
+app_logger.addHandler(console_handler)
+
+async def run_process(disable_console: bool, rate_unit: str, run_error_case: bool):
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        total_steps = 100
+        tracker = ProgressTracker(total=total_steps, description="Simulating a process...", rate_unit=rate_unit)
+        all_logs = []
+        last_log_content = None
+        
+        initial_log = f"Starting simulated process with {total_steps} steps..."
+        app_logger.info(initial_log)
+        logs = [
+            {
+                "type": "log",
+                "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
+                "content": record.getMessage()
+            }
+            for record in get_logs()
+        ]
+        all_logs.extend(logs)
+        last_log_content = logs[-1]["content"] if logs else None        
+        yield tracker.update(advance=0, status="running", logs=all_logs, log_content=None)
+
+        for i in range(total_steps):
+            await asyncio.sleep(0.03)
+            current_step = i + 1
+            
+            if current_step == 10:
+                app_logger.warning(f"Low disk space warning at step {current_step}.")
+            elif current_step == 30:
+                app_logger.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
+            elif current_step == 75:
+                app_logger.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
+            
+            if run_error_case and current_step == 50:
+                app_logger.error("A fatal simulation error occurred! Aborting.")
+                logs = [
+                    {
+                        "type": "log",
+                        "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
+                        "content": record.getMessage()
+                    }
+                    for record in get_logs()
+                ]
+                all_logs.extend(logs)
+                last_log_content = logs[-1]["content"] if logs else last_log_content
+                yield tracker.update(advance=0, status="error", logs=all_logs, log_content=last_log_content)
+                return
+            
+            logs = [
+                {
+                    "type": "log",
+                    "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
+                    "content": record.getMessage()
+                }
+                for record in get_logs()
+            ]
+            all_logs.extend(logs)
+            if logs:
+                last_log_content = logs[-1]["content"]
+            yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
+        
+        final_log = "Process completed successfully!"
+        app_logger.log(logging.INFO + 5, final_log)
+        logs = [
+            {
+                "type": "log",
+                "level": "SUCCESS" if record.levelno == logging.INFO + 5 else record.levelname,
+                "content": record.getMessage()
+            }
+            for record in get_logs()
+        ]
+        all_logs.extend(logs)
+        last_log_content = logs[-1]["content"] if logs else last_log_content
+        yield tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
+        
+def update_livelog_properties(mode, color, lines, scroll):
+    return gr.update(display_mode=mode, background_color=color, line_numbers=lines, autoscroll=scroll)
+
+def clear_output():
+    return None
+
+async def run_success_case(disable_console: bool, rate_unit: str):
+    yield None    
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=False):
+        yield update
+
+async def run_error_case(disable_console: bool, rate_unit: str):
+    yield None
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=True):
+        yield update
+
+
+# --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
+diffusion_pipeline = None
+pipeline_lock = threading.Lock()
+def load_pipeline(on_load=True):
+    """A function to load the model, ensuring it's only done once."""
+    global diffusion_pipeline
+    with pipeline_lock:
+        if diffusion_pipeline is None:
+            print("Loading Stable Diffusion model for the first time...")
+            pipe = StableDiffusionXLPipeline.from_pretrained(
+                MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
+            )
+            pipe.enable_vae_tiling()
+            pipe.enable_model_cpu_offload()
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)        
+            diffusion_pipeline = pipe
+            print("Model loaded successfully!")
+        
+    if not on_load:
+        return diffusion_pipeline
+
+@spaces.GPU(duration=60, enable_queue=True)
+def run_diffusion_in_thread(prompt: str, disable_console: bool, update_queue: queue.Queue):
+    """
+    This function now uses capture_logs to listen to internal diffusers logs
+    while retaining the superior data structure you designed.
+    """
+    tracker = None    
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        try:            
+            pipe = load_pipeline(on_load=False)
+            
+            #We will capture pipeline tqdm s/it progress instead            
+            rate_queue = queue.Queue()
+            tqdm_writer = TqdmToQueueWriter(rate_queue)
+            
+            progress_bar_handler = Tee(sys.stderr, tqdm_writer)
+            pipe.set_progress_bar_config(file=progress_bar_handler,  #if you dont need to see the tqdm progress on console set file=tqdm_writer instead            
+                                        disable=False,  
+                                        ncols=100,
+                                        dynamic_ncols=True,
+                                        ascii=" █")
+            
+            seed = random.randint(0, MAX_SEED)
+            generator = torch.Generator(device="cuda").manual_seed(seed)
+            prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
+            negative_prompt_style = "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly"
+            num_inference_steps = 10
+            
+            all_logs = []
+            last_known_rate_data = None
+
+            # Helper function to process and store new logs
+            def process_and_send_updates(status="running", advance=0, final_image_payload=None):
+                """
+                This is the core callback function. It captures new logs, formats them,
+                and sends a complete update object (logs + progress) to the UI queue.
+                This should also be called after the log record.
+                """
+                nonlocal all_logs, last_known_rate_data
+                new_rate_data = None
+                while not rate_queue.empty():
+                    try:
+                        new_rate_data = rate_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                
+                if new_rate_data is not None:
+                    last_known_rate_data = new_rate_data
+                
+                new_records = get_logs()
+                if new_records:
+                    new_logs = [{
+                        "type": "log",
+                        "level": "SUCCESS" if r.levelno == logging.INFO + 5 else r.levelname,
+                        "content": r.getMessage()
+                    } for r in new_records]
+                    all_logs.extend(new_logs)
+                
+                # Use the tracker to generate the progress update dictionary if it exists.
+                # If not, create a preliminary update dictionary.
+                update_dict = {}
+                
+                if tracker:
+                    update_dict = tracker.update(
+                        advance=advance, 
+                        status=status, 
+                        logs=all_logs,
+                        rate_data=last_known_rate_data                               
+                    )
+                else:
+                    # Initial state before the tracker is created.
+                    update_dict = {
+                        "type": "progress", 
+                        "logs": all_logs, 
+                        "current": 0, 
+                        "total": num_inference_steps, 
+                        "desc": "Diffusion Steps" # Description is sent once
+                    }
+
+                # Put the update on the queue. The image payload is usually None
+                # until the very end.
+                update_queue.put((final_image_payload, update_dict))
+                
+            app_logger.info(f"Using seed: {seed}")
+            process_and_send_updates()
+                        
+            app_logger.info("Starting diffusion process...")
+            process_and_send_updates()
+                        
+            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps", rate_unit='it/s')
+            
+            def progress_callback(pipe_instance, step, timestep, callback_kwargs):
+                process_and_send_updates(advance=1) 
+                return callback_kwargs
+                        
+            images = pipe(
+                prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
+                guidance_scale=3.0, num_inference_steps=num_inference_steps,
+                generator=generator, callback_on_step_end=progress_callback
+            ).images
+            
+            app_logger.log(logging.INFO + 5, "Image generated successfully!")
+            process_and_send_updates(status="success", final_image_payload=images)
+           
+
+        except Exception as e:
+            app_logger.error(f"Error in diffusion thread: {e}, process aborted!", exc_info=True)                    
+            process_and_send_updates(status="error")                                                        
+        finally:
+            update_queue.put(None)
+            
+            
+@spaces.GPU(duration=60, enable_queue=True)
+def generate(prompt):
+    """This function starts the worker thread and yields updates from the queue."""   
+    yield None, None, gr.update(interactive=False)    
+    update_queue = queue.Queue()
+    diffusion_thread = threading.Thread(target=run_diffusion_in_thread,  args=(prompt, False, update_queue))
+    diffusion_thread.start()
+    final_images = None
+    while True:
+        update = update_queue.get()
+        if update is None: 
+            break
+        
+        images, log_update = update
+        
+        if images:
+            final_images = images
+      
+        yield final_images, log_update, gr.skip()
+    
+    yield final_images, log_update, gr.update(interactive=True)
+
+# --- 4. THE COMBINED GRADIO UI with TABS ---
+with gr.Blocks(theme=gr.themes.Ocean()) as demo:
+    gr.HTML("<h1><center>LiveLog Component Showcase</center></h1>")
+
+    with gr.Tabs():
+        with gr.TabItem("LiveLog Feature Demo"):            
+            gr.Markdown("### Test all features of the LiveLog component interactively.")
+            with gr.Row():
+                with gr.Column(scale=3):
+                    feature_logger = LiveLog(
+                        label="Process Output", line_numbers=True, height=550,
+                        background_color="#000000", display_mode="full"
+                    )
+                with gr.Column(scale=1):
+                    with gr.Group():
+                        gr.Markdown("### Component Properties")
+                        display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
+                        rate_unit = gr.Radio(["it/s","s/it"], label="Progress rate unit", value="it/s")
+                        bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
+                        line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
+                        autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
+                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
+                    with gr.Group():
+                        gr.Markdown("### Simulation Controls")
+                        start_btn = gr.Button("Run Success Case", variant="primary")
+                        error_btn = gr.Button("Run Error Case")
+            
+            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
+            controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
+            for control in controls:
+                control.change(fn=update_livelog_properties, inputs=controls, outputs=feature_logger)
+        
+        with gr.TabItem("Diffusion Pipeline Integration"):               
+            gr.Markdown("### Use `LiveLog` to monitor a real image generation process.")
+            with gr.Row():
+                with gr.Column(scale=3):
+                    with gr.Group():
+                        prompt = gr.Textbox(
+                            label="Enter your prompt", show_label=False,
+                            placeholder="A cinematic photo of a robot in a floral garden...",
+                            scale=8, container=False
+                        )
+                        run_button = gr.Button("Generate", scale=1, variant="primary")
+                    
+                    livelog_viewer = LiveLog(
+                        label="Process Monitor", height=250, display_mode="full", line_numbers=False
+                    )
+                
+                with gr.Column(scale=2):
+                    result_gallery = gr.Gallery(
+                        label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
+                    )
+            
+            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer, run_button])
+            prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
+            livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
+            
+    # This ensures the model is downloaded/loaded once when the app starts.
+    demo.load(load_pipeline, None, None)
+                
+if __name__ == "__main__":
     demo.queue(max_size=50).launch(debug=True)

requirements.txt

@@ -1,8 +1,9 @@
-diffusers
-safetensors
-transformers
-accelerate
-gradio_livelog>=0.0.2
-torch
-torchvision
-torchaudio
+--extra-index-url https://download.pytorch.org/whl/cu124
+diffusers
+safetensors
+transformers
+accelerate
+gradio_livelog
+torch==2.6.0+cu124
+torchvision==0.21.0+cu124
+torchaudio==2.6.0+cu124

space.py

@@ -9,7 +9,7 @@ abs_path = os.path.join(os.path.dirname(__file__), "css.css")
 
 with gr.Blocks(
     css=abs_path,
-    theme=gr.themes.Ocean(
+    theme=gr.themes.Default(
         font_mono=[
             gr.themes.GoogleFont("Inconsolata"),
             "monospace",
@@ -40,9 +40,9 @@ pip install gradio_livelog
 ```python
 # demo/app.py
 
+import sys
 import gradio as gr
 import torch
-import time
 import logging
 import random
 import numpy as np
@@ -54,24 +54,28 @@ import spaces
 
 # Import the component and ALL its utilities
 from gradio_livelog import LiveLog
-from gradio_livelog.utils import ProgressTracker, capture_logs
+from gradio_livelog.utils import ProgressTracker, Tee, TqdmToQueueWriter, capture_logs
 
 # --- 1. SETUP ---
 MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
 MAX_SEED = np.iinfo(np.int32).max
 
 # --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
-logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
-
-async def run_process(disable_console: bool, run_error_case: bool):
-    with capture_logs(disable_console=disable_console) as get_logs:
+app_logger = logging.getLogger("logging_app")
+app_logger.setLevel(logging.INFO)
+console_handler = logging.StreamHandler()
+console_handler.flush = sys.stderr.flush
+app_logger.addHandler(console_handler)
+
+async def run_process(disable_console: bool, rate_unit: str, run_error_case: bool):
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
         total_steps = 100
-        tracker = ProgressTracker(total=total_steps, description="Simulating a process...")
+        tracker = ProgressTracker(total=total_steps, description="Simulating a process...", rate_unit=rate_unit)
         all_logs = []
         last_log_content = None
         
         initial_log = f"Starting simulated process with {total_steps} steps..."
-        logging.info(initial_log)
+        app_logger.info(initial_log)
         logs = [
             {
                 "type": "log",
@@ -89,14 +93,14 @@ async def run_process(disable_console: bool, run_error_case: bool):
             current_step = i + 1
             
             if current_step == 10:
-                logging.warning(f"Low disk space warning at step {current_step}.")
+                app_logger.warning(f"Low disk space warning at step {current_step}.")
             elif current_step == 30:
-                logging.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
+                app_logger.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
             elif current_step == 75:
-                logging.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
+                app_logger.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
             
             if run_error_case and current_step == 50:
-                logging.error("A fatal simulation error occurred! Aborting.")
+                app_logger.error("A fatal simulation error occurred! Aborting.")
                 logs = [
                     {
                         "type": "log",
@@ -124,7 +128,7 @@ async def run_process(disable_console: bool, run_error_case: bool):
             yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
         
         final_log = "Process completed successfully!"
-        logging.log(logging.INFO + 5, final_log)
+        app_logger.log(logging.INFO + 5, final_log)
         logs = [
             {
                 "type": "log",
@@ -143,47 +147,60 @@ def update_livelog_properties(mode, color, lines, scroll):
 def clear_output():
     return None
 
-async def run_success_case(disable_console: bool):
+async def run_success_case(disable_console: bool, rate_unit: str):
     yield None    
-    async for update in run_process(disable_console=disable_console, run_error_case=False):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=False):
         yield update
 
-async def run_error_case(disable_console: bool):
+async def run_error_case(disable_console: bool, rate_unit: str):
     yield None
-    async for update in run_process(disable_console=disable_console, run_error_case=True):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=True):
         yield update
 
 
 # --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
 diffusion_pipeline = None
+pipeline_lock = threading.Lock()
 def load_pipeline(on_load=True):
     \"\"\"A function to load the model, ensuring it's only done once.\"\"\"
     global diffusion_pipeline
-    if diffusion_pipeline is None:
-        print("Loading Stable Diffusion model for the first time...")
-        pipe = StableDiffusionXLPipeline.from_pretrained(
-            MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
-        )
-        pipe.enable_vae_tiling()
-        pipe.enable_model_cpu_offload()
-        pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-        pipe.set_progress_bar_config(disable=True)
-        diffusion_pipeline = pipe
-        print("Model loaded successfully.")
-    
+    with pipeline_lock:
+        if diffusion_pipeline is None:
+            print("Loading Stable Diffusion model for the first time...")
+            pipe = StableDiffusionXLPipeline.from_pretrained(
+                MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
+            )
+            pipe.enable_vae_tiling()
+            pipe.enable_model_cpu_offload()
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)        
+            diffusion_pipeline = pipe
+            print("Model loaded successfully!")
+        
     if not on_load:
         return diffusion_pipeline
 
 @spaces.GPU(duration=60, enable_queue=True)
-def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
+def run_diffusion_in_thread(prompt: str, disable_console: bool, update_queue: queue.Queue):
     \"\"\"
     This function now uses capture_logs to listen to internal diffusers logs
     while retaining the superior data structure you designed.
     \"\"\"
     tracker = None    
-    with capture_logs() as get_logs:
-        try:
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        try:            
             pipe = load_pipeline(on_load=False)
+            
+            #We will capture pipeline tqdm s/it progress instead            
+            rate_queue = queue.Queue()
+            tqdm_writer = TqdmToQueueWriter(rate_queue)
+            
+            progress_bar_handler = Tee(sys.stderr, tqdm_writer)
+            pipe.set_progress_bar_config(file=progress_bar_handler,  #if you dont need to see the tqdm progress on console set file=tqdm_writer instead            
+                                        disable=False,  
+                                        ncols=100,
+                                        dynamic_ncols=True,
+                                        ascii=" █")
+            
             seed = random.randint(0, MAX_SEED)
             generator = torch.Generator(device="cuda").manual_seed(seed)
             prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
@@ -191,67 +208,110 @@ def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
             num_inference_steps = 10
             
             all_logs = []
-            last_log_content = None
+            last_known_rate_data = None
 
             # Helper function to process and store new logs
-            def process_and_store_logs():
-                nonlocal all_logs, last_log_content
+            def process_and_send_updates(status="running", advance=0, final_image_payload=None):
+                \"\"\"
+                This is the core callback function. It captures new logs, formats them,
+                and sends a complete update object (logs + progress) to the UI queue.
+                This should also be called after the log record.
+                \"\"\"
+                nonlocal all_logs, last_known_rate_data
+                new_rate_data = None
+                while not rate_queue.empty():
+                    try:
+                        new_rate_data = rate_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                
+                if new_rate_data is not None:
+                    last_known_rate_data = new_rate_data
+                
                 new_records = get_logs()
                 if new_records:
-                    new_logs = [{"type": "log", "level": r.levelname, "content": r.getMessage()} for r in new_records]
+                    new_logs = [{
+                        "type": "log",
+                        "level": "SUCCESS" if r.levelno == logging.INFO + 5 else r.levelname,
+                        "content": r.getMessage()
+                    } for r in new_records]
                     all_logs.extend(new_logs)
-                    last_log_content = all_logs[-1]["content"]
-            
-            logging.info(f"Using seed: {seed}")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            logging.info("Starting diffusion process...")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps")
+                
+                # Use the tracker to generate the progress update dictionary if it exists.
+                # If not, create a preliminary update dictionary.
+                update_dict = {}
+                
+                if tracker:
+                    update_dict = tracker.update(
+                        advance=advance, 
+                        status=status, 
+                        logs=all_logs,
+                        rate_data=last_known_rate_data                               
+                    )
+                else:
+                    # Initial state before the tracker is created.
+                    update_dict = {
+                        "type": "progress", 
+                        "logs": all_logs, 
+                        "current": 0, 
+                        "total": num_inference_steps, 
+                        "desc": "Diffusion Steps" # Description is sent once
+                    }
+
+                # Put the update on the queue. The image payload is usually None
+                # until the very end.
+                update_queue.put((final_image_payload, update_dict))
+                
+            app_logger.info(f"Using seed: {seed}")
+            process_and_send_updates()
+                        
+            app_logger.info("Starting diffusion process...")
+            process_and_send_updates()
+                        
+            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps", rate_unit='it/s')
             
             def progress_callback(pipe_instance, step, timestep, callback_kwargs):
-                process_and_store_logs() # Check for new logs from diffusers at each step
-                update_dict = tracker.update(logs=all_logs)
-                update_queue.put((None, update_dict))
+                process_and_send_updates(advance=1) 
                 return callback_kwargs
-            
+                        
             images = pipe(
                 prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
                 guidance_scale=3.0, num_inference_steps=num_inference_steps,
                 generator=generator, callback_on_step_end=progress_callback
             ).images
             
-            logging.log(logging.INFO + 5, "Image generated successfully!")
-            process_and_store_logs()
-            
-            final_update = tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-            update_queue.put((images, final_update))
+            app_logger.log(logging.INFO + 5, "Image generated successfully!")
+            process_and_send_updates(status="success", final_image_payload=images)
+           
 
         except Exception as e:
-            logging.error(f"Error in diffusion thread: {e}", exc_info=True)
-            process_and_store_logs() # Capture the final error log
-            if tracker:
-                error_update = tracker.update(advance=0, status="error", logs=all_logs, log_content=f"An error occurred: {e}")
-                update_queue.put((None, error_update))
+            app_logger.error(f"Error in diffusion thread: {e}, process aborted!", exc_info=True)                    
+            process_and_send_updates(status="error")                                                        
         finally:
             update_queue.put(None)
             
+            
 @spaces.GPU(duration=60, enable_queue=True)
 def generate(prompt):
     \"\"\"This function starts the worker thread and yields updates from the queue.\"\"\"   
-    yield None, None
-    yield None, {"type": "log", "level": "INFO", "content": "Preparing generation..."}
+    yield None, None, gr.update(interactive=False)    
     update_queue = queue.Queue()
-    diffusion_thread = threading.Thread(target=run_diffusion_in_thread, args=(prompt, update_queue))
+    diffusion_thread = threading.Thread(target=run_diffusion_in_thread,  args=(prompt, False, update_queue))
     diffusion_thread.start()
+    final_images = None
     while True:
         update = update_queue.get()
-        if update is None: break
-        yield update
-
+        if update is None: 
+            break
+        
+        images, log_update = update
+        
+        if images:
+            final_images = images
+      
+        yield final_images, log_update, gr.skip()
+    
+    yield final_images, log_update, gr.update(interactive=True)
 
 # --- 4. THE COMBINED GRADIO UI with TABS ---
 with gr.Blocks(theme=gr.themes.Ocean()) as demo:
@@ -270,17 +330,18 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                     with gr.Group():
                         gr.Markdown("### Component Properties")
                         display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
+                        rate_unit = gr.Radio(["it/s","s/it"], label="Progress rate unit", value="it/s")
                         bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
                         line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
                         autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
+                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                     with gr.Group():
                         gr.Markdown("### Simulation Controls")
-                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                         start_btn = gr.Button("Run Success Case", variant="primary")
                         error_btn = gr.Button("Run Error Case")
             
-            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox], outputs=feature_logger)
+            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
             feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
             controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
             for control in controls:
@@ -307,7 +368,7 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                         label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
                     )
             
-            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
+            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer, run_button])
             prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
             livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
             

src/README.md

@@ -10,23 +10,33 @@ app_file: space.py
 ---
 
 # `gradio_livelog`
-<img alt="Static Badge" src="https://img.shields.io/badge/version%20-%200.0.1%20-%20blue"> <a href="https://huggingface.co/spaces/elismasilva/gradio_livelog"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Demo-blue"></a><p><span>💻 <a href='https://github.com/DEVAIEXP/gradio_component_livelog'>Component GitHub Code</a></span></p>
+<img alt="Static Badge" src="https://img.shields.io/badge/version%20-%200.0.3%20-%20blue"> <a href="https://huggingface.co/spaces/elismasilva/gradio_livelog"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Demo-blue"></a><p><span>💻 <a href='https://github.com/DEVAIEXP/gradio_component_livelog'>Component GitHub Code</a></span></p>
 
 A Live Log Component for Gradio Interface
 
 ## Key Features
 
-`LiveLog` provides a rich, terminal-like experience directly in your Gradio UI.
+`LiveLog` elevates Gradio applications by providing a powerful, terminal-like monitoring experience directly in your UI. It's designed for both simple progress tracking and complex pipeline introspection.
 
-- **Real-time Log Streaming:** Display log messages as they are generated in your Python backend, perfect for monitoring long-running tasks.
-- **Integrated Progress Bar:** Track the progress of loops and processes with a familiar, `tqdm`-style progress bar. 
-    - Includes iterations per second (it/s) calculation.
-    - The bar automatically changes color to green on success or red on error for clear visual feedback.
-- **Automatic `logging` Capture:** Effortlessly capture all logs generated by Python's standard `logging` module (`logging.info`, `logging.error`, etc.) using the `capture_logs` context manager.
-- **Flexible Display Modes:** Choose to display the full component (logs + progress bar), logs only, or the progress bar only to fit different UI layouts.
-- **Highly Customizable:** Tailor the component's appearance with properties for background color, line numbers, and autoscrolling.
-- **Utility Controls:** Comes with built-in buttons to clear the log, copy all content to the clipboard, and download logs as a text file.
-- **Console Silencing:** Optionally suppress log output in your Python console to keep it clean while still displaying everything in the UI.
+-   **Dual-Mode Progress Tracking:** A sophisticated progress bar that operates in two modes for maximum accuracy:
+    -   **Internal Rate Calculation:** For simple loops, it features a built-in, `tqdm`-style progress calculator with **Exponential Moving Average (EMA)** smoothing for a stable and realistic `it/s` or `s/it` display.
+    -   **External `tqdm` Capture:** For deep integration, it can **directly capture and display the *exact* rate** from an existing `tqdm` instance running inside a backend library (like `diffusers`). This eliminates measurement overhead and provides a perfectly synchronized view of your pipeline's true performance.
+
+-   **Rich, Real-time Log Streaming:** Display log messages as they are generated.
+    -   Supports standard Python log levels (`INFO`, `WARNING`, `ERROR`, etc.) with corresponding colors.
+    -   Includes support for **custom log levels** (like "SUCCESS") for enhanced visual feedback.
+
+-   **Advanced Multi-Logger Capture:** The `capture_logs` utility is designed for complex applications.
+    -   Effortlessly capture logs from **multiple, independent Python loggers** simultaneously (e.g., your app's logger and a library's internal logger).
+    -   Correctly handles logger hierarchies and propagation, making it robust for any logging setup.
+
+-   **Flexible Display & Layout Control:** Adapt the component to any UI layout.
+    -   **Three Display Modes:** Show the full component (`logs + progress`), `logs only`, or `progress bar only`.
+    -   Highly customizable appearance with properties for `height`, `background_color`, `line_numbers`, and `autoscrolling`.
+
+-   **Comprehensive Utility Controls:**
+    -   Built-in header buttons to **Clear**, **Copy**, and **Download** log content.
+    -   Optionally suppress log output in your Python console to keep it clean while still displaying everything in the UI.
 
 ## Installation
 
@@ -39,9 +49,9 @@ pip install gradio_livelog
 ```python
 # demo/app.py
 
+import sys
 import gradio as gr
 import torch
-import time
 import logging
 import random
 import numpy as np
@@ -53,24 +63,28 @@ import spaces
 
 # Import the component and ALL its utilities
 from gradio_livelog import LiveLog
-from gradio_livelog.utils import ProgressTracker, capture_logs
+from gradio_livelog.utils import ProgressTracker, Tee, TqdmToQueueWriter, capture_logs
 
 # --- 1. SETUP ---
 MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
 MAX_SEED = np.iinfo(np.int32).max
 
 # --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
-logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
-
-async def run_process(disable_console: bool, run_error_case: bool):
-    with capture_logs(disable_console=disable_console) as get_logs:
+app_logger = logging.getLogger("logging_app")
+app_logger.setLevel(logging.INFO)
+console_handler = logging.StreamHandler()
+console_handler.flush = sys.stderr.flush
+app_logger.addHandler(console_handler)
+
+async def run_process(disable_console: bool, rate_unit: str, run_error_case: bool):
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
         total_steps = 100
-        tracker = ProgressTracker(total=total_steps, description="Simulating a process...")
+        tracker = ProgressTracker(total=total_steps, description="Simulating a process...", rate_unit=rate_unit)
         all_logs = []
         last_log_content = None
         
         initial_log = f"Starting simulated process with {total_steps} steps..."
-        logging.info(initial_log)
+        app_logger.info(initial_log)
         logs = [
             {
                 "type": "log",
@@ -88,14 +102,14 @@ async def run_process(disable_console: bool, run_error_case: bool):
             current_step = i + 1
             
             if current_step == 10:
-                logging.warning(f"Low disk space warning at step {current_step}.")
+                app_logger.warning(f"Low disk space warning at step {current_step}.")
             elif current_step == 30:
-                logging.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
+                app_logger.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
             elif current_step == 75:
-                logging.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
+                app_logger.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
             
             if run_error_case and current_step == 50:
-                logging.error("A fatal simulation error occurred! Aborting.")
+                app_logger.error("A fatal simulation error occurred! Aborting.")
                 logs = [
                     {
                         "type": "log",
@@ -123,7 +137,7 @@ async def run_process(disable_console: bool, run_error_case: bool):
             yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
         
         final_log = "Process completed successfully!"
-        logging.log(logging.INFO + 5, final_log)
+        app_logger.log(logging.INFO + 5, final_log)
         logs = [
             {
                 "type": "log",
@@ -142,47 +156,60 @@ def update_livelog_properties(mode, color, lines, scroll):
 def clear_output():
     return None
 
-async def run_success_case(disable_console: bool):
+async def run_success_case(disable_console: bool, rate_unit: str):
     yield None    
-    async for update in run_process(disable_console=disable_console, run_error_case=False):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=False):
         yield update
 
-async def run_error_case(disable_console: bool):
+async def run_error_case(disable_console: bool, rate_unit: str):
     yield None
-    async for update in run_process(disable_console=disable_console, run_error_case=True):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=True):
         yield update
 
 
 # --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
 diffusion_pipeline = None
+pipeline_lock = threading.Lock()
 def load_pipeline(on_load=True):
     """A function to load the model, ensuring it's only done once."""
     global diffusion_pipeline
-    if diffusion_pipeline is None:
-        print("Loading Stable Diffusion model for the first time...")
-        pipe = StableDiffusionXLPipeline.from_pretrained(
-            MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
-        )
-        pipe.enable_vae_tiling()
-        pipe.enable_model_cpu_offload()
-        pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-        pipe.set_progress_bar_config(disable=True)
-        diffusion_pipeline = pipe
-        print("Model loaded successfully.")
-    
+    with pipeline_lock:
+        if diffusion_pipeline is None:
+            print("Loading Stable Diffusion model for the first time...")
+            pipe = StableDiffusionXLPipeline.from_pretrained(
+                MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
+            )
+            pipe.enable_vae_tiling()
+            pipe.enable_model_cpu_offload()
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)        
+            diffusion_pipeline = pipe
+            print("Model loaded successfully!")
+        
     if not on_load:
         return diffusion_pipeline
 
 @spaces.GPU(duration=60, enable_queue=True)
-def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
+def run_diffusion_in_thread(prompt: str, disable_console: bool, update_queue: queue.Queue):
     """
     This function now uses capture_logs to listen to internal diffusers logs
     while retaining the superior data structure you designed.
     """
     tracker = None    
-    with capture_logs() as get_logs:
-        try:
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        try:            
             pipe = load_pipeline(on_load=False)
+            
+            #We will capture pipeline tqdm s/it progress instead            
+            rate_queue = queue.Queue()
+            tqdm_writer = TqdmToQueueWriter(rate_queue)
+            
+            progress_bar_handler = Tee(sys.stderr, tqdm_writer)
+            pipe.set_progress_bar_config(file=progress_bar_handler,  #if you dont need to see the tqdm progress on console set file=tqdm_writer instead            
+                                        disable=False,  
+                                        ncols=100,
+                                        dynamic_ncols=True,
+                                        ascii=" █")
+            
             seed = random.randint(0, MAX_SEED)
             generator = torch.Generator(device="cuda").manual_seed(seed)
             prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
@@ -190,67 +217,110 @@ def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
             num_inference_steps = 10
             
             all_logs = []
-            last_log_content = None
+            last_known_rate_data = None
 
             # Helper function to process and store new logs
-            def process_and_store_logs():
-                nonlocal all_logs, last_log_content
+            def process_and_send_updates(status="running", advance=0, final_image_payload=None):
+                """
+                This is the core callback function. It captures new logs, formats them,
+                and sends a complete update object (logs + progress) to the UI queue.
+                This should also be called after the log record.
+                """
+                nonlocal all_logs, last_known_rate_data
+                new_rate_data = None
+                while not rate_queue.empty():
+                    try:
+                        new_rate_data = rate_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                
+                if new_rate_data is not None:
+                    last_known_rate_data = new_rate_data
+                
                 new_records = get_logs()
                 if new_records:
-                    new_logs = [{"type": "log", "level": r.levelname, "content": r.getMessage()} for r in new_records]
+                    new_logs = [{
+                        "type": "log",
+                        "level": "SUCCESS" if r.levelno == logging.INFO + 5 else r.levelname,
+                        "content": r.getMessage()
+                    } for r in new_records]
                     all_logs.extend(new_logs)
-                    last_log_content = all_logs[-1]["content"]
-            
-            logging.info(f"Using seed: {seed}")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            logging.info("Starting diffusion process...")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps")
+                
+                # Use the tracker to generate the progress update dictionary if it exists.
+                # If not, create a preliminary update dictionary.
+                update_dict = {}
+                
+                if tracker:
+                    update_dict = tracker.update(
+                        advance=advance, 
+                        status=status, 
+                        logs=all_logs,
+                        rate_data=last_known_rate_data                               
+                    )
+                else:
+                    # Initial state before the tracker is created.
+                    update_dict = {
+                        "type": "progress", 
+                        "logs": all_logs, 
+                        "current": 0, 
+                        "total": num_inference_steps, 
+                        "desc": "Diffusion Steps" # Description is sent once
+                    }
+
+                # Put the update on the queue. The image payload is usually None
+                # until the very end.
+                update_queue.put((final_image_payload, update_dict))
+                
+            app_logger.info(f"Using seed: {seed}")
+            process_and_send_updates()
+                        
+            app_logger.info("Starting diffusion process...")
+            process_and_send_updates()
+                        
+            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps", rate_unit='it/s')
             
             def progress_callback(pipe_instance, step, timestep, callback_kwargs):
-                process_and_store_logs() # Check for new logs from diffusers at each step
-                update_dict = tracker.update(logs=all_logs)
-                update_queue.put((None, update_dict))
+                process_and_send_updates(advance=1) 
                 return callback_kwargs
-            
+                        
             images = pipe(
                 prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
                 guidance_scale=3.0, num_inference_steps=num_inference_steps,
                 generator=generator, callback_on_step_end=progress_callback
             ).images
             
-            logging.log(logging.INFO + 5, "Image generated successfully!")
-            process_and_store_logs()
-            
-            final_update = tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-            update_queue.put((images, final_update))
+            app_logger.log(logging.INFO + 5, "Image generated successfully!")
+            process_and_send_updates(status="success", final_image_payload=images)
+           
 
         except Exception as e:
-            logging.error(f"Error in diffusion thread: {e}", exc_info=True)
-            process_and_store_logs() # Capture the final error log
-            if tracker:
-                error_update = tracker.update(advance=0, status="error", logs=all_logs, log_content=f"An error occurred: {e}")
-                update_queue.put((None, error_update))
+            app_logger.error(f"Error in diffusion thread: {e}, process aborted!", exc_info=True)                    
+            process_and_send_updates(status="error")                                                        
         finally:
             update_queue.put(None)
             
+            
 @spaces.GPU(duration=60, enable_queue=True)
 def generate(prompt):
     """This function starts the worker thread and yields updates from the queue."""   
-    yield None, None
-    yield None, {"type": "log", "level": "INFO", "content": "Preparing generation..."}
+    yield None, None, gr.update(interactive=False)    
     update_queue = queue.Queue()
-    diffusion_thread = threading.Thread(target=run_diffusion_in_thread, args=(prompt, update_queue))
+    diffusion_thread = threading.Thread(target=run_diffusion_in_thread,  args=(prompt, False, update_queue))
     diffusion_thread.start()
+    final_images = None
     while True:
         update = update_queue.get()
-        if update is None: break
-        yield update
-
+        if update is None: 
+            break
+        
+        images, log_update = update
+        
+        if images:
+            final_images = images
+      
+        yield final_images, log_update, gr.skip()
+    
+    yield final_images, log_update, gr.update(interactive=True)
 
 # --- 4. THE COMBINED GRADIO UI with TABS ---
 with gr.Blocks(theme=gr.themes.Ocean()) as demo:
@@ -269,17 +339,18 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                     with gr.Group():
                         gr.Markdown("### Component Properties")
                         display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
+                        rate_unit = gr.Radio(["it/s","s/it"], label="Progress rate unit", value="it/s")
                         bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
                         line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
                         autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
+                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                     with gr.Group():
                         gr.Markdown("### Simulation Controls")
-                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                         start_btn = gr.Button("Run Success Case", variant="primary")
                         error_btn = gr.Button("Run Error Case")
             
-            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox], outputs=feature_logger)
+            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
             feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
             controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
             for control in controls:
@@ -306,7 +377,7 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                         label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
                     )
             
-            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
+            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer, run_button])
             prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
             livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
             

src/backend/gradio_livelog/templates/component/style.css

@@ -1 +1 @@
-.block.svelte-239wnu{position:relative;margin:0;box-shadow:var(--block-shadow);border-width:var(--block-border-width);border-color:var(--block-border-color);border-radius:var(--block-radius);background:var(--block-background-fill);width:100%;line-height:var(--line-sm)}.block.fullscreen.svelte-239wnu{border-radius:0}.auto-margin.svelte-239wnu{margin-left:auto;margin-right:auto}.block.border_focus.svelte-239wnu{border-color:var(--color-accent)}.block.border_contrast.svelte-239wnu{border-color:var(--body-text-color)}.padded.svelte-239wnu{padding:var(--block-padding)}.hidden.svelte-239wnu{display:none}.flex.svelte-239wnu{display:flex;flex-direction:column}.hide-container.svelte-239wnu:not(.fullscreen){margin:0;box-shadow:none;--block-border-width:0;background:transparent;padding:0;overflow:visible}.resize-handle.svelte-239wnu{position:absolute;bottom:0;right:0;width:10px;height:10px;fill:var(--block-border-color);cursor:nwse-resize}.fullscreen.svelte-239wnu{position:fixed;top:0;left:0;width:100vw;height:100vh;z-index:1000;overflow:auto}.animating.svelte-239wnu{animation:svelte-239wnu-pop-out .1s ease-out forwards}@keyframes svelte-239wnu-pop-out{0%{position:fixed;top:var(--start-top);left:var(--start-left);width:var(--start-width);height:var(--start-height);z-index:100}to{position:fixed;top:0vh;left:0vw;width:100vw;height:100vh;z-index:1000}}.placeholder.svelte-239wnu{border-radius:var(--block-radius);border-width:var(--block-border-width);border-color:var(--block-border-color);border-style:dashed}Tables */ table,tr,td,th{margin-top:var(--spacing-sm);margin-bottom:var(--spacing-sm);padding:var(--spacing-xl)}.md code,.md pre{background:none;font-family:var(--font-mono);font-size:var(--text-sm);text-align:left;white-space:pre;word-spacing:normal;word-break:normal;word-wrap:normal;line-height:1.5;-moz-tab-size:2;tab-size:2;-webkit-hyphens:none;hyphens:none}.md pre[class*=language-]::selection,.md pre[class*=language-] ::selection,.md code[class*=language-]::selection,.md code[class*=language-] ::selection{text-shadow:none;background:#b3d4fc}.md pre{padding:1em;margin:.5em 0;overflow:auto;position:relative;margin-top:var(--spacing-sm);margin-bottom:var(--spacing-sm);box-shadow:none;border:none;border-radius:var(--radius-md);background:var(--code-background-fill);padding:var(--spacing-xxl);font-family:var(--font-mono);text-shadow:none;border-radius:var(--radius-sm);white-space:nowrap;display:block;white-space:pre}.md :not(pre)>code{padding:.1em;border-radius:var(--radius-xs);white-space:normal;background:var(--code-background-fill);border:1px solid var(--panel-border-color);padding:var(--spacing-xxs) var(--spacing-xs)}.md .token.comment,.md .token.prolog,.md .token.doctype,.md .token.cdata{color:#708090}.md .token.punctuation{color:#999}.md .token.namespace{opacity:.7}.md .token.property,.md .token.tag,.md .token.boolean,.md .token.number,.md .token.constant,.md .token.symbol,.md .token.deleted{color:#905}.md .token.selector,.md .token.attr-name,.md .token.string,.md .token.char,.md .token.builtin,.md .token.inserted{color:#690}.md .token.atrule,.md .token.attr-value,.md .token.keyword{color:#07a}.md .token.function,.md .token.class-name{color:#dd4a68}.md .token.regex,.md .token.important,.md .token.variable{color:#e90}.md .token.important,.md .token.bold{font-weight:700}.md .token.italic{font-style:italic}.md .token.entity{cursor:help}.dark .md .token.comment,.dark .md .token.prolog,.dark .md .token.cdata{color:#5c6370}.dark .md .token.doctype,.dark .md .token.punctuation,.dark .md .token.entity{color:#abb2bf}.dark .md .token.attr-name,.dark .md .token.class-name,.dark .md .token.boolean,.dark .md .token.constant,.dark .md .token.number,.dark .md .token.atrule{color:#d19a66}.dark .md .token.keyword{color:#c678dd}.dark .md .token.property,.dark .md .token.tag,.dark .md .token.symbol,.dark .md .token.deleted,.dark .md .token.important{color:#e06c75}.dark .md .token.selector,.dark .md .token.string,.dark .md .token.char,.dark .md .token.builtin,.dark .md .token.inserted,.dark .md .token.regex,.dark .md .token.attr-value,.dark .md .token.attr-value>.token.punctuation{color:#98c379}.dark .md .token.variable,.dark .md .token.operator,.dark .md .token.function{color:#61afef}.dark .md .token.url{color:#56b6c2}span.svelte-1m32c2s div[class*=code_wrap]{position:relative}span.svelte-1m32c2s span.katex{font-size:var(--text-lg);direction:ltr}span.svelte-1m32c2s div[class*=code_wrap]>button{z-index:1;cursor:pointer;border-bottom-left-radius:var(--radius-sm);padding:var(--spacing-md);width:25px;height:25px;position:absolute;right:0}span.svelte-1m32c2s .check{opacity:0;z-index:var(--layer-top);transition:opacity .2s;background:var(--code-background-fill);color:var(--body-text-color);position:absolute;top:var(--size-1-5);left:var(--size-1-5)}span.svelte-1m32c2s p:not(:first-child){margin-top:var(--spacing-xxl)}span.svelte-1m32c2s .md-header-anchor{margin-left:-25px;padding-right:8px;line-height:1;color:var(--body-text-color-subdued);opacity:0}span.svelte-1m32c2s h1:hover .md-header-anchor,span.svelte-1m32c2s h2:hover .md-header-anchor,span.svelte-1m32c2s h3:hover .md-header-anchor,span.svelte-1m32c2s h4:hover .md-header-anchor,span.svelte-1m32c2s h5:hover .md-header-anchor,span.svelte-1m32c2s h6:hover .md-header-anchor{opacity:1}span.md.svelte-1m32c2s .md-header-anchor>svg{color:var(--body-text-color-subdued)}span.svelte-1m32c2s table{word-break:break-word}div.svelte-17qq50w>.md.prose{font-weight:var(--block-info-text-weight);font-size:var(--block-info-text-size);line-height:var(--line-sm)}div.svelte-17qq50w>.md.prose *{color:var(--block-info-text-color)}div.svelte-17qq50w{margin-bottom:var(--spacing-md)}span.has-info.svelte-zgrq3{margin-bottom:var(--spacing-xs)}span.svelte-zgrq3:not(.has-info){margin-bottom:var(--spacing-lg)}span.svelte-zgrq3{display:inline-block;position:relative;z-index:var(--layer-4);border:solid var(--block-title-border-width) var(--block-title-border-color);border-radius:var(--block-title-radius);background:var(--block-title-background-fill);padding:var(--block-title-padding);color:var(--block-title-text-color);font-weight:var(--block-title-text-weight);font-size:var(--block-title-text-size);line-height:var(--line-sm)}span[dir=rtl].svelte-zgrq3{display:block}.hide.svelte-zgrq3{margin:0;height:0}label.svelte-13ao5pu.svelte-13ao5pu{display:inline-flex;align-items:center;z-index:var(--layer-2);box-shadow:var(--block-label-shadow);border:var(--block-label-border-width) solid var(--block-label-border-color);border-top:none;border-left:none;border-radius:var(--block-label-radius);background:var(--block-label-background-fill);padding:var(--block-label-padding);pointer-events:none;color:var(--block-label-text-color);font-weight:var(--block-label-text-weight);font-size:var(--block-label-text-size);line-height:var(--line-sm)}.gr-group label.svelte-13ao5pu.svelte-13ao5pu{border-top-left-radius:0}label.float.svelte-13ao5pu.svelte-13ao5pu{position:absolute;top:var(--block-label-margin);left:var(--block-label-margin)}label.svelte-13ao5pu.svelte-13ao5pu:not(.float){position:static;margin-top:var(--block-label-margin);margin-left:var(--block-label-margin)}.hide.svelte-13ao5pu.svelte-13ao5pu{height:0}span.svelte-13ao5pu.svelte-13ao5pu{opacity:.8;margin-right:var(--size-2);width:calc(var(--block-label-text-size) - 1px);height:calc(var(--block-label-text-size) - 1px)}.hide-label.svelte-13ao5pu.svelte-13ao5pu{box-shadow:none;border-width:0;background:transparent;overflow:visible}label[dir=rtl].svelte-13ao5pu.svelte-13ao5pu{border:var(--block-label-border-width) solid var(--block-label-border-color);border-top:none;border-right:none;border-bottom-left-radius:var(--block-radius);border-bottom-right-radius:var(--block-label-radius);border-top-left-radius:var(--block-label-radius)}label[dir=rtl].svelte-13ao5pu span.svelte-13ao5pu{margin-left:var(--size-2);margin-right:0}button.svelte-qgco6m{display:flex;justify-content:center;align-items:center;gap:1px;z-index:var(--layer-2);border-radius:var(--radius-xs);color:var(--block-label-text-color);border:1px solid transparent;padding:var(--spacing-xxs)}button.svelte-qgco6m:hover{background-color:var(--background-fill-secondary)}button[disabled].svelte-qgco6m{opacity:.5;box-shadow:none}button[disabled].svelte-qgco6m:hover{cursor:not-allowed}.padded.svelte-qgco6m{background:var(--bg-color)}button.svelte-qgco6m:hover,button.highlight.svelte-qgco6m{cursor:pointer;color:var(--color-accent)}.padded.svelte-qgco6m:hover{color:var(--block-label-text-color)}span.svelte-qgco6m{padding:0 1px;font-size:10px}div.svelte-qgco6m{display:flex;align-items:center;justify-content:center;transition:filter .2s ease-in-out}.x-small.svelte-qgco6m{width:10px;height:10px}.small.svelte-qgco6m{width:14px;height:14px}.medium.svelte-qgco6m{width:20px;height:20px}.large.svelte-qgco6m{width:22px;height:22px}.pending.svelte-qgco6m{animation:svelte-qgco6m-flash .5s infinite}@keyframes svelte-qgco6m-flash{0%{opacity:.5}50%{opacity:1}to{opacity:.5}}.transparent.svelte-qgco6m{background:transparent;border:none;box-shadow:none}.empty.svelte-3w3rth{display:flex;justify-content:center;align-items:center;margin-top:calc(0px - var(--size-6));height:var(--size-full)}.icon.svelte-3w3rth{opacity:.5;height:var(--size-5);color:var(--body-text-color)}.small.svelte-3w3rth{min-height:calc(var(--size-32) - 20px)}.large.svelte-3w3rth{min-height:calc(var(--size-64) - 20px)}.unpadded_box.svelte-3w3rth{margin-top:0}.small_parent.svelte-3w3rth{min-height:100%!important}.dropdown-arrow.svelte-145leq6,.dropdown-arrow.svelte-ihhdbf{fill:currentColor}.circle.svelte-ihhdbf{fill:currentColor;opacity:.1}svg.svelte-pb9pol{animation:svelte-pb9pol-spin 1.5s linear infinite}@keyframes svelte-pb9pol-spin{0%{transform:rotate(0)}to{transform:rotate(360deg)}}h2.svelte-1xg7h5n{font-size:var(--text-xl)!important}p.svelte-1xg7h5n,h2.svelte-1xg7h5n{white-space:pre-line}.wrap.svelte-1xg7h5n{display:flex;flex-direction:column;justify-content:center;align-items:center;min-height:var(--size-60);color:var(--block-label-text-color);line-height:var(--line-md);height:100%;padding-top:var(--size-3);text-align:center;margin:auto var(--spacing-lg)}.or.svelte-1xg7h5n{color:var(--body-text-color-subdued);display:flex}.icon-wrap.svelte-1xg7h5n{width:30px;margin-bottom:var(--spacing-lg)}@media (--screen-md){.wrap.svelte-1xg7h5n{font-size:var(--text-lg)}}.hovered.svelte-1xg7h5n{color:var(--color-accent)}div.svelte-q32hvf{border-top:1px solid transparent;display:flex;max-height:100%;justify-content:center;align-items:center;gap:var(--spacing-sm);height:auto;align-items:flex-end;color:var(--block-label-text-color);flex-shrink:0}.show_border.svelte-q32hvf{border-top:1px solid var(--block-border-color);margin-top:var(--spacing-xxl);box-shadow:var(--shadow-drop)}.source-selection.svelte-15ls1gu{display:flex;align-items:center;justify-content:center;border-top:1px solid var(--border-color-primary);width:100%;margin-left:auto;margin-right:auto;height:var(--size-10)}.icon.svelte-15ls1gu{width:22px;height:22px;margin:var(--spacing-lg) var(--spacing-xs);padding:var(--spacing-xs);color:var(--neutral-400);border-radius:var(--radius-md)}.selected.svelte-15ls1gu{color:var(--color-accent)}.icon.svelte-15ls1gu:hover,.icon.svelte-15ls1gu:focus{color:var(--color-accent)}.icon-button-wrapper.svelte-109se4{display:flex;flex-direction:row;align-items:center;justify-content:center;z-index:var(--layer-3);gap:var(--spacing-sm);box-shadow:var(--shadow-drop);border:1px solid var(--border-color-primary);background:var(--block-background-fill);padding:var(--spacing-xxs)}.icon-button-wrapper.hide-top-corner.svelte-109se4{border-top:none;border-right:none;border-radius:var(--block-label-right-radius)}.icon-button-wrapper.display-top-corner.svelte-109se4{border-radius:var(--radius-sm) 0 0 var(--radius-sm);top:var(--spacing-sm);right:-1px}.icon-button-wrapper.svelte-109se4:not(.top-panel){border:1px solid var(--border-color-primary);border-radius:var(--radius-sm)}.top-panel.svelte-109se4{position:absolute;top:var(--block-label-margin);right:var(--block-label-margin);margin:0}.icon-button-wrapper.svelte-109se4 button{margin:var(--spacing-xxs);border-radius:var(--radius-xs);position:relative}.icon-button-wrapper.svelte-109se4 a.download-link:not(:last-child),.icon-button-wrapper.svelte-109se4 button:not(:last-child){margin-right:var(--spacing-xxs)}.icon-button-wrapper.svelte-109se4 a.download-link:not(:last-child):not(.no-border *):after,.icon-button-wrapper.svelte-109se4 button:not(:last-child):not(.no-border *):after{content:"";position:absolute;right:-4.5px;top:15%;height:70%;width:1px;background-color:var(--border-color-primary)}.icon-button-wrapper.svelte-109se4>*{height:100%}svg.svelte-43sxxs.svelte-43sxxs{width:var(--size-20);height:var(--size-20)}svg.svelte-43sxxs path.svelte-43sxxs{fill:var(--loader-color)}div.svelte-43sxxs.svelte-43sxxs{z-index:var(--layer-2)}.margin.svelte-43sxxs.svelte-43sxxs{margin:var(--size-4)}.wrap.svelte-17v219f.svelte-17v219f{display:flex;flex-direction:column;justify-content:center;align-items:center;z-index:var(--layer-2);transition:opacity .1s ease-in-out;border-radius:var(--block-radius);background:var(--block-background-fill);padding:0 var(--size-6);max-height:var(--size-screen-h);overflow:hidden}.wrap.center.svelte-17v219f.svelte-17v219f{top:0;right:0;left:0}.wrap.default.svelte-17v219f.svelte-17v219f{top:0;right:0;bottom:0;left:0}.hide.svelte-17v219f.svelte-17v219f{opacity:0;pointer-events:none}.generating.svelte-17v219f.svelte-17v219f{animation:svelte-17v219f-pulseStart 1s cubic-bezier(.4,0,.6,1),svelte-17v219f-pulse 2s cubic-bezier(.4,0,.6,1) 1s infinite;border:2px solid var(--color-accent);background:transparent;z-index:var(--layer-1);pointer-events:none}.translucent.svelte-17v219f.svelte-17v219f{background:none}@keyframes svelte-17v219f-pulseStart{0%{opacity:0}to{opacity:1}}@keyframes svelte-17v219f-pulse{0%,to{opacity:1}50%{opacity:.5}}.loading.svelte-17v219f.svelte-17v219f{z-index:var(--layer-2);color:var(--body-text-color)}.eta-bar.svelte-17v219f.svelte-17v219f{position:absolute;top:0;right:0;bottom:0;left:0;transform-origin:left;opacity:.8;z-index:var(--layer-1);transition:10ms;background:var(--background-fill-secondary)}.progress-bar-wrap.svelte-17v219f.svelte-17v219f{border:1px solid var(--border-color-primary);background:var(--background-fill-primary);width:55.5%;height:var(--size-4)}.progress-bar.svelte-17v219f.svelte-17v219f{transform-origin:left;background-color:var(--loader-color);width:var(--size-full);height:var(--size-full)}.progress-level.svelte-17v219f.svelte-17v219f{display:flex;flex-direction:column;align-items:center;gap:1;z-index:var(--layer-2);width:var(--size-full)}.progress-level-inner.svelte-17v219f.svelte-17v219f{margin:var(--size-2) auto;color:var(--body-text-color);font-size:var(--text-sm);font-family:var(--font-mono)}.meta-text.svelte-17v219f.svelte-17v219f{position:absolute;bottom:0;right:0;z-index:var(--layer-2);padding:var(--size-1) var(--size-2);font-size:var(--text-sm);font-family:var(--font-mono)}.meta-text-center.svelte-17v219f.svelte-17v219f{display:flex;position:absolute;top:0;right:0;justify-content:center;align-items:center;transform:translateY(var(--size-6));z-index:var(--layer-2);padding:var(--size-1) var(--size-2);font-size:var(--text-sm);font-family:var(--font-mono);text-align:center}.error.svelte-17v219f.svelte-17v219f{box-shadow:var(--shadow-drop);border:solid 1px var(--error-border-color);border-radius:var(--radius-full);background:var(--error-background-fill);padding-right:var(--size-4);padding-left:var(--size-4);color:var(--error-text-color);font-weight:var(--weight-semibold);font-size:var(--text-lg);line-height:var(--line-lg);font-family:var(--font)}.minimal.svelte-17v219f.svelte-17v219f{pointer-events:none}.minimal.svelte-17v219f .progress-text.svelte-17v219f{background:var(--block-background-fill)}.border.svelte-17v219f.svelte-17v219f{border:1px solid var(--border-color-primary)}.clear-status.svelte-17v219f.svelte-17v219f{position:absolute;display:flex;top:var(--size-2);right:var(--size-2);justify-content:flex-end;gap:var(--spacing-sm);z-index:var(--layer-1)}.toast-body.svelte-syezpc{display:flex;position:relative;right:0;left:0;align-items:center;margin:var(--size-6) var(--size-4);margin:auto;border-radius:var(--container-radius);overflow:hidden;pointer-events:auto}.toast-body.error.svelte-syezpc{border:1px solid var(--color-red-700);background:var(--color-red-50)}.dark .toast-body.error.svelte-syezpc{border:1px solid var(--color-red-500);background-color:var(--color-grey-950)}.toast-body.warning.svelte-syezpc{border:1px solid var(--color-yellow-700);background:var(--color-yellow-50)}.dark .toast-body.warning.svelte-syezpc{border:1px solid var(--color-yellow-500);background-color:var(--color-grey-950)}.toast-body.info.svelte-syezpc{border:1px solid var(--color-grey-700);background:var(--color-grey-50)}.dark .toast-body.info.svelte-syezpc{border:1px solid var(--color-grey-500);background-color:var(--color-grey-950)}.toast-body.success.svelte-syezpc{border:1px solid var(--color-green-700);background:var(--color-green-50)}.dark .toast-body.success.svelte-syezpc{border:1px solid var(--color-green-500);background-color:var(--color-grey-950)}.toast-title.svelte-syezpc{display:flex;align-items:center;font-weight:var(--weight-bold);font-size:var(--text-lg);line-height:var(--line-sm)}.toast-title.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-title.error.svelte-syezpc{color:var(--color-red-50)}.toast-title.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-title.warning.svelte-syezpc{color:var(--color-yellow-50)}.toast-title.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-title.info.svelte-syezpc{color:var(--color-grey-50)}.toast-title.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-title.success.svelte-syezpc{color:var(--color-green-50)}.toast-close.svelte-syezpc{margin:0 var(--size-3);border-radius:var(--size-3);padding:0px var(--size-1-5);font-size:var(--size-5);line-height:var(--size-5)}.toast-close.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-close.error.svelte-syezpc{color:var(--color-red-500)}.toast-close.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-close.warning.svelte-syezpc{color:var(--color-yellow-500)}.toast-close.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-close.info.svelte-syezpc{color:var(--color-grey-500)}.toast-close.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-close.success.svelte-syezpc{color:var(--color-green-500)}.toast-text.svelte-syezpc{font-size:var(--text-lg);word-wrap:break-word;overflow-wrap:break-word;word-break:break-word}.toast-text.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-text.error.svelte-syezpc{color:var(--color-red-50)}.toast-text.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-text.warning.svelte-syezpc{color:var(--color-yellow-50)}.toast-text.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-text.info.svelte-syezpc{color:var(--color-grey-50)}.toast-text.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-text.success.svelte-syezpc{color:var(--color-green-50)}.toast-details.svelte-syezpc{margin:var(--size-3) var(--size-3) var(--size-3) 0;width:100%}.toast-icon.svelte-syezpc{display:flex;position:absolute;position:relative;flex-shrink:0;justify-content:center;align-items:center;margin:var(--size-2);border-radius:var(--radius-full);padding:var(--size-1);padding-left:calc(var(--size-1) - 1px);width:35px;height:35px}.toast-icon.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-icon.error.svelte-syezpc{color:var(--color-red-500)}.toast-icon.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-icon.warning.svelte-syezpc{color:var(--color-yellow-500)}.toast-icon.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-icon.info.svelte-syezpc{color:var(--color-grey-500)}.toast-icon.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-icon.success.svelte-syezpc{color:var(--color-green-500)}@keyframes svelte-syezpc-countdown{0%{transform:scaleX(1)}to{transform:scaleX(0)}}.timer.svelte-syezpc{position:absolute;bottom:0;left:0;transform-origin:0 0;animation:svelte-syezpc-countdown 10s linear forwards;width:100%;height:var(--size-1)}.timer.error.svelte-syezpc{background:var(--color-red-700)}.dark .timer.error.svelte-syezpc{background:var(--color-red-500)}.timer.warning.svelte-syezpc{background:var(--color-yellow-700)}.dark .timer.warning.svelte-syezpc{background:var(--color-yellow-500)}.timer.info.svelte-syezpc{background:var(--color-grey-700)}.dark .timer.info.svelte-syezpc{background:var(--color-grey-500)}.timer.success.svelte-syezpc{background:var(--color-green-700)}.dark .timer.success.svelte-syezpc{background:var(--color-green-500)}.hidden.svelte-syezpc{display:none}.toast-text.svelte-syezpc a{text-decoration:underline}.toast-wrap.svelte-gatr8h{display:flex;position:fixed;top:var(--size-4);right:var(--size-4);flex-direction:column;align-items:end;gap:var(--size-2);z-index:var(--layer-top);width:calc(100% - var(--size-8))}@media (--screen-sm){.toast-wrap.svelte-gatr8h{width:calc(var(--size-96) + var(--size-10))}}.streaming-bar.svelte-ga0jj6{position:absolute;bottom:0;left:0;right:0;height:4px;background-color:var(--primary-600);animation:svelte-ga0jj6-countdown linear forwards;z-index:1}@keyframes svelte-ga0jj6-countdown{0%{transform:translate(0)}to{transform:translate(-100%)}}.unstyled-link.svelte-151nsdd{all:unset;cursor:pointer}.panel-container.svelte-fg67bo{display:flex;flex-direction:column;border:1px solid var(--border-color-primary);border-radius:0!important;background-color:var(--background-fill-primary);overflow:hidden}.log-view-container.svelte-fg67bo{display:flex;flex-direction:column;flex-grow:1;min-height:0}.header.svelte-fg67bo{border-bottom:1px solid var(--border-color-primary);background-color:var(--background-fill-secondary);display:flex;justify-content:flex-end;flex-shrink:0}.log-panel.svelte-fg67bo{flex-grow:1;font-family:var(--font-mono, monospace);font-size:var(--text-sm);overflow-y:auto;color:#f8f8f8}.log-line.svelte-fg67bo{display:flex}.line-number.svelte-fg67bo{opacity:.6;padding-right:var(--spacing-lg);-webkit-user-select:none;user-select:none;text-align:right;min-width:3ch}.log-content.svelte-fg67bo{margin:0;padding-left:5px;white-space:pre-wrap;word-break:break-word}.log-level-info.svelte-fg67bo{color:inherit}.log-level-debug.svelte-fg67bo{color:#888}.log-level-warning.svelte-fg67bo{color:#facc15}.log-level-error.svelte-fg67bo{color:#ef4444}.log-level-critical.svelte-fg67bo{background-color:#ef4444;color:#fff;font-weight:700;padding:0 .25rem}.log-level-success.svelte-fg67bo{color:#22c55e}.progress-container.svelte-fg67bo{padding:var(--spacing-sm) var(--spacing-md);border-top:1px solid var(--border-color-primary);background:var(--background-fill-secondary)}.progress-label-top.svelte-fg67bo,.progress-label-bottom.svelte-fg67bo{display:flex;justify-content:space-between;font-size:var(--text-sm);color:var(--body-text-color-subdued)}.progress-label-top.svelte-fg67bo{margin-bottom:var(--spacing-xs)}.progress-label-bottom.svelte-fg67bo{margin-top:var(--spacing-xs)}.progress-bar-background.svelte-fg67bo{width:100%;height:8px;background-color:var(--background-fill-primary);border-radius:var(--radius-full);overflow:hidden}.progress-bar-fill.svelte-fg67bo{height:100%;background-color:var(--color-accent);border-radius:var(--radius-full);transition:width .1s linear,background-color .3s ease}.progress-bar-fill.success.svelte-fg67bo{background-color:var(--color-success, #22c55e)}.progress-bar-fill.error.svelte-fg67bo{background-color:var(--color-error, #ef4444)}.block-label-wrapper.svelte-10ojysx{padding-bottom:24px}
+.block.svelte-239wnu{position:relative;margin:0;box-shadow:var(--block-shadow);border-width:var(--block-border-width);border-color:var(--block-border-color);border-radius:var(--block-radius);background:var(--block-background-fill);width:100%;line-height:var(--line-sm)}.block.fullscreen.svelte-239wnu{border-radius:0}.auto-margin.svelte-239wnu{margin-left:auto;margin-right:auto}.block.border_focus.svelte-239wnu{border-color:var(--color-accent)}.block.border_contrast.svelte-239wnu{border-color:var(--body-text-color)}.padded.svelte-239wnu{padding:var(--block-padding)}.hidden.svelte-239wnu{display:none}.flex.svelte-239wnu{display:flex;flex-direction:column}.hide-container.svelte-239wnu:not(.fullscreen){margin:0;box-shadow:none;--block-border-width:0;background:transparent;padding:0;overflow:visible}.resize-handle.svelte-239wnu{position:absolute;bottom:0;right:0;width:10px;height:10px;fill:var(--block-border-color);cursor:nwse-resize}.fullscreen.svelte-239wnu{position:fixed;top:0;left:0;width:100vw;height:100vh;z-index:1000;overflow:auto}.animating.svelte-239wnu{animation:svelte-239wnu-pop-out .1s ease-out forwards}@keyframes svelte-239wnu-pop-out{0%{position:fixed;top:var(--start-top);left:var(--start-left);width:var(--start-width);height:var(--start-height);z-index:100}to{position:fixed;top:0vh;left:0vw;width:100vw;height:100vh;z-index:1000}}.placeholder.svelte-239wnu{border-radius:var(--block-radius);border-width:var(--block-border-width);border-color:var(--block-border-color);border-style:dashed}Tables */ table,tr,td,th{margin-top:var(--spacing-sm);margin-bottom:var(--spacing-sm);padding:var(--spacing-xl)}.md code,.md pre{background:none;font-family:var(--font-mono);font-size:var(--text-sm);text-align:left;white-space:pre;word-spacing:normal;word-break:normal;word-wrap:normal;line-height:1.5;-moz-tab-size:2;tab-size:2;-webkit-hyphens:none;hyphens:none}.md pre[class*=language-]::selection,.md pre[class*=language-] ::selection,.md code[class*=language-]::selection,.md code[class*=language-] ::selection{text-shadow:none;background:#b3d4fc}.md pre{padding:1em;margin:.5em 0;overflow:auto;position:relative;margin-top:var(--spacing-sm);margin-bottom:var(--spacing-sm);box-shadow:none;border:none;border-radius:var(--radius-md);background:var(--code-background-fill);padding:var(--spacing-xxl);font-family:var(--font-mono);text-shadow:none;border-radius:var(--radius-sm);white-space:nowrap;display:block;white-space:pre}.md :not(pre)>code{padding:.1em;border-radius:var(--radius-xs);white-space:normal;background:var(--code-background-fill);border:1px solid var(--panel-border-color);padding:var(--spacing-xxs) var(--spacing-xs)}.md .token.comment,.md .token.prolog,.md .token.doctype,.md .token.cdata{color:#708090}.md .token.punctuation{color:#999}.md .token.namespace{opacity:.7}.md .token.property,.md .token.tag,.md .token.boolean,.md .token.number,.md .token.constant,.md .token.symbol,.md .token.deleted{color:#905}.md .token.selector,.md .token.attr-name,.md .token.string,.md .token.char,.md .token.builtin,.md .token.inserted{color:#690}.md .token.atrule,.md .token.attr-value,.md .token.keyword{color:#07a}.md .token.function,.md .token.class-name{color:#dd4a68}.md .token.regex,.md .token.important,.md .token.variable{color:#e90}.md .token.important,.md .token.bold{font-weight:700}.md .token.italic{font-style:italic}.md .token.entity{cursor:help}.dark .md .token.comment,.dark .md .token.prolog,.dark .md .token.cdata{color:#5c6370}.dark .md .token.doctype,.dark .md .token.punctuation,.dark .md .token.entity{color:#abb2bf}.dark .md .token.attr-name,.dark .md .token.class-name,.dark .md .token.boolean,.dark .md .token.constant,.dark .md .token.number,.dark .md .token.atrule{color:#d19a66}.dark .md .token.keyword{color:#c678dd}.dark .md .token.property,.dark .md .token.tag,.dark .md .token.symbol,.dark .md .token.deleted,.dark .md .token.important{color:#e06c75}.dark .md .token.selector,.dark .md .token.string,.dark .md .token.char,.dark .md .token.builtin,.dark .md .token.inserted,.dark .md .token.regex,.dark .md .token.attr-value,.dark .md .token.attr-value>.token.punctuation{color:#98c379}.dark .md .token.variable,.dark .md .token.operator,.dark .md .token.function{color:#61afef}.dark .md .token.url{color:#56b6c2}span.svelte-1m32c2s div[class*=code_wrap]{position:relative}span.svelte-1m32c2s span.katex{font-size:var(--text-lg);direction:ltr}span.svelte-1m32c2s div[class*=code_wrap]>button{z-index:1;cursor:pointer;border-bottom-left-radius:var(--radius-sm);padding:var(--spacing-md);width:25px;height:25px;position:absolute;right:0}span.svelte-1m32c2s .check{opacity:0;z-index:var(--layer-top);transition:opacity .2s;background:var(--code-background-fill);color:var(--body-text-color);position:absolute;top:var(--size-1-5);left:var(--size-1-5)}span.svelte-1m32c2s p:not(:first-child){margin-top:var(--spacing-xxl)}span.svelte-1m32c2s .md-header-anchor{margin-left:-25px;padding-right:8px;line-height:1;color:var(--body-text-color-subdued);opacity:0}span.svelte-1m32c2s h1:hover .md-header-anchor,span.svelte-1m32c2s h2:hover .md-header-anchor,span.svelte-1m32c2s h3:hover .md-header-anchor,span.svelte-1m32c2s h4:hover .md-header-anchor,span.svelte-1m32c2s h5:hover .md-header-anchor,span.svelte-1m32c2s h6:hover .md-header-anchor{opacity:1}span.md.svelte-1m32c2s .md-header-anchor>svg{color:var(--body-text-color-subdued)}span.svelte-1m32c2s table{word-break:break-word}div.svelte-17qq50w>.md.prose{font-weight:var(--block-info-text-weight);font-size:var(--block-info-text-size);line-height:var(--line-sm)}div.svelte-17qq50w>.md.prose *{color:var(--block-info-text-color)}div.svelte-17qq50w{margin-bottom:var(--spacing-md)}span.has-info.svelte-zgrq3{margin-bottom:var(--spacing-xs)}span.svelte-zgrq3:not(.has-info){margin-bottom:var(--spacing-lg)}span.svelte-zgrq3{display:inline-block;position:relative;z-index:var(--layer-4);border:solid var(--block-title-border-width) var(--block-title-border-color);border-radius:var(--block-title-radius);background:var(--block-title-background-fill);padding:var(--block-title-padding);color:var(--block-title-text-color);font-weight:var(--block-title-text-weight);font-size:var(--block-title-text-size);line-height:var(--line-sm)}span[dir=rtl].svelte-zgrq3{display:block}.hide.svelte-zgrq3{margin:0;height:0}label.svelte-13ao5pu.svelte-13ao5pu{display:inline-flex;align-items:center;z-index:var(--layer-2);box-shadow:var(--block-label-shadow);border:var(--block-label-border-width) solid var(--block-label-border-color);border-top:none;border-left:none;border-radius:var(--block-label-radius);background:var(--block-label-background-fill);padding:var(--block-label-padding);pointer-events:none;color:var(--block-label-text-color);font-weight:var(--block-label-text-weight);font-size:var(--block-label-text-size);line-height:var(--line-sm)}.gr-group label.svelte-13ao5pu.svelte-13ao5pu{border-top-left-radius:0}label.float.svelte-13ao5pu.svelte-13ao5pu{position:absolute;top:var(--block-label-margin);left:var(--block-label-margin)}label.svelte-13ao5pu.svelte-13ao5pu:not(.float){position:static;margin-top:var(--block-label-margin);margin-left:var(--block-label-margin)}.hide.svelte-13ao5pu.svelte-13ao5pu{height:0}span.svelte-13ao5pu.svelte-13ao5pu{opacity:.8;margin-right:var(--size-2);width:calc(var(--block-label-text-size) - 1px);height:calc(var(--block-label-text-size) - 1px)}.hide-label.svelte-13ao5pu.svelte-13ao5pu{box-shadow:none;border-width:0;background:transparent;overflow:visible}label[dir=rtl].svelte-13ao5pu.svelte-13ao5pu{border:var(--block-label-border-width) solid var(--block-label-border-color);border-top:none;border-right:none;border-bottom-left-radius:var(--block-radius);border-bottom-right-radius:var(--block-label-radius);border-top-left-radius:var(--block-label-radius)}label[dir=rtl].svelte-13ao5pu span.svelte-13ao5pu{margin-left:var(--size-2);margin-right:0}button.svelte-qgco6m{display:flex;justify-content:center;align-items:center;gap:1px;z-index:var(--layer-2);border-radius:var(--radius-xs);color:var(--block-label-text-color);border:1px solid transparent;padding:var(--spacing-xxs)}button.svelte-qgco6m:hover{background-color:var(--background-fill-secondary)}button[disabled].svelte-qgco6m{opacity:.5;box-shadow:none}button[disabled].svelte-qgco6m:hover{cursor:not-allowed}.padded.svelte-qgco6m{background:var(--bg-color)}button.svelte-qgco6m:hover,button.highlight.svelte-qgco6m{cursor:pointer;color:var(--color-accent)}.padded.svelte-qgco6m:hover{color:var(--block-label-text-color)}span.svelte-qgco6m{padding:0 1px;font-size:10px}div.svelte-qgco6m{display:flex;align-items:center;justify-content:center;transition:filter .2s ease-in-out}.x-small.svelte-qgco6m{width:10px;height:10px}.small.svelte-qgco6m{width:14px;height:14px}.medium.svelte-qgco6m{width:20px;height:20px}.large.svelte-qgco6m{width:22px;height:22px}.pending.svelte-qgco6m{animation:svelte-qgco6m-flash .5s infinite}@keyframes svelte-qgco6m-flash{0%{opacity:.5}50%{opacity:1}to{opacity:.5}}.transparent.svelte-qgco6m{background:transparent;border:none;box-shadow:none}.empty.svelte-3w3rth{display:flex;justify-content:center;align-items:center;margin-top:calc(0px - var(--size-6));height:var(--size-full)}.icon.svelte-3w3rth{opacity:.5;height:var(--size-5);color:var(--body-text-color)}.small.svelte-3w3rth{min-height:calc(var(--size-32) - 20px)}.large.svelte-3w3rth{min-height:calc(var(--size-64) - 20px)}.unpadded_box.svelte-3w3rth{margin-top:0}.small_parent.svelte-3w3rth{min-height:100%!important}.dropdown-arrow.svelte-145leq6,.dropdown-arrow.svelte-ihhdbf{fill:currentColor}.circle.svelte-ihhdbf{fill:currentColor;opacity:.1}svg.svelte-pb9pol{animation:svelte-pb9pol-spin 1.5s linear infinite}@keyframes svelte-pb9pol-spin{0%{transform:rotate(0)}to{transform:rotate(360deg)}}h2.svelte-1xg7h5n{font-size:var(--text-xl)!important}p.svelte-1xg7h5n,h2.svelte-1xg7h5n{white-space:pre-line}.wrap.svelte-1xg7h5n{display:flex;flex-direction:column;justify-content:center;align-items:center;min-height:var(--size-60);color:var(--block-label-text-color);line-height:var(--line-md);height:100%;padding-top:var(--size-3);text-align:center;margin:auto var(--spacing-lg)}.or.svelte-1xg7h5n{color:var(--body-text-color-subdued);display:flex}.icon-wrap.svelte-1xg7h5n{width:30px;margin-bottom:var(--spacing-lg)}@media (--screen-md){.wrap.svelte-1xg7h5n{font-size:var(--text-lg)}}.hovered.svelte-1xg7h5n{color:var(--color-accent)}div.svelte-q32hvf{border-top:1px solid transparent;display:flex;max-height:100%;justify-content:center;align-items:center;gap:var(--spacing-sm);height:auto;align-items:flex-end;color:var(--block-label-text-color);flex-shrink:0}.show_border.svelte-q32hvf{border-top:1px solid var(--block-border-color);margin-top:var(--spacing-xxl);box-shadow:var(--shadow-drop)}.source-selection.svelte-15ls1gu{display:flex;align-items:center;justify-content:center;border-top:1px solid var(--border-color-primary);width:100%;margin-left:auto;margin-right:auto;height:var(--size-10)}.icon.svelte-15ls1gu{width:22px;height:22px;margin:var(--spacing-lg) var(--spacing-xs);padding:var(--spacing-xs);color:var(--neutral-400);border-radius:var(--radius-md)}.selected.svelte-15ls1gu{color:var(--color-accent)}.icon.svelte-15ls1gu:hover,.icon.svelte-15ls1gu:focus{color:var(--color-accent)}.icon-button-wrapper.svelte-109se4{display:flex;flex-direction:row;align-items:center;justify-content:center;z-index:var(--layer-3);gap:var(--spacing-sm);box-shadow:var(--shadow-drop);border:1px solid var(--border-color-primary);background:var(--block-background-fill);padding:var(--spacing-xxs)}.icon-button-wrapper.hide-top-corner.svelte-109se4{border-top:none;border-right:none;border-radius:var(--block-label-right-radius)}.icon-button-wrapper.display-top-corner.svelte-109se4{border-radius:var(--radius-sm) 0 0 var(--radius-sm);top:var(--spacing-sm);right:-1px}.icon-button-wrapper.svelte-109se4:not(.top-panel){border:1px solid var(--border-color-primary);border-radius:var(--radius-sm)}.top-panel.svelte-109se4{position:absolute;top:var(--block-label-margin);right:var(--block-label-margin);margin:0}.icon-button-wrapper.svelte-109se4 button{margin:var(--spacing-xxs);border-radius:var(--radius-xs);position:relative}.icon-button-wrapper.svelte-109se4 a.download-link:not(:last-child),.icon-button-wrapper.svelte-109se4 button:not(:last-child){margin-right:var(--spacing-xxs)}.icon-button-wrapper.svelte-109se4 a.download-link:not(:last-child):not(.no-border *):after,.icon-button-wrapper.svelte-109se4 button:not(:last-child):not(.no-border *):after{content:"";position:absolute;right:-4.5px;top:15%;height:70%;width:1px;background-color:var(--border-color-primary)}.icon-button-wrapper.svelte-109se4>*{height:100%}svg.svelte-43sxxs.svelte-43sxxs{width:var(--size-20);height:var(--size-20)}svg.svelte-43sxxs path.svelte-43sxxs{fill:var(--loader-color)}div.svelte-43sxxs.svelte-43sxxs{z-index:var(--layer-2)}.margin.svelte-43sxxs.svelte-43sxxs{margin:var(--size-4)}.wrap.svelte-17v219f.svelte-17v219f{display:flex;flex-direction:column;justify-content:center;align-items:center;z-index:var(--layer-2);transition:opacity .1s ease-in-out;border-radius:var(--block-radius);background:var(--block-background-fill);padding:0 var(--size-6);max-height:var(--size-screen-h);overflow:hidden}.wrap.center.svelte-17v219f.svelte-17v219f{top:0;right:0;left:0}.wrap.default.svelte-17v219f.svelte-17v219f{top:0;right:0;bottom:0;left:0}.hide.svelte-17v219f.svelte-17v219f{opacity:0;pointer-events:none}.generating.svelte-17v219f.svelte-17v219f{animation:svelte-17v219f-pulseStart 1s cubic-bezier(.4,0,.6,1),svelte-17v219f-pulse 2s cubic-bezier(.4,0,.6,1) 1s infinite;border:2px solid var(--color-accent);background:transparent;z-index:var(--layer-1);pointer-events:none}.translucent.svelte-17v219f.svelte-17v219f{background:none}@keyframes svelte-17v219f-pulseStart{0%{opacity:0}to{opacity:1}}@keyframes svelte-17v219f-pulse{0%,to{opacity:1}50%{opacity:.5}}.loading.svelte-17v219f.svelte-17v219f{z-index:var(--layer-2);color:var(--body-text-color)}.eta-bar.svelte-17v219f.svelte-17v219f{position:absolute;top:0;right:0;bottom:0;left:0;transform-origin:left;opacity:.8;z-index:var(--layer-1);transition:10ms;background:var(--background-fill-secondary)}.progress-bar-wrap.svelte-17v219f.svelte-17v219f{border:1px solid var(--border-color-primary);background:var(--background-fill-primary);width:55.5%;height:var(--size-4)}.progress-bar.svelte-17v219f.svelte-17v219f{transform-origin:left;background-color:var(--loader-color);width:var(--size-full);height:var(--size-full)}.progress-level.svelte-17v219f.svelte-17v219f{display:flex;flex-direction:column;align-items:center;gap:1;z-index:var(--layer-2);width:var(--size-full)}.progress-level-inner.svelte-17v219f.svelte-17v219f{margin:var(--size-2) auto;color:var(--body-text-color);font-size:var(--text-sm);font-family:var(--font-mono)}.meta-text.svelte-17v219f.svelte-17v219f{position:absolute;bottom:0;right:0;z-index:var(--layer-2);padding:var(--size-1) var(--size-2);font-size:var(--text-sm);font-family:var(--font-mono)}.meta-text-center.svelte-17v219f.svelte-17v219f{display:flex;position:absolute;top:0;right:0;justify-content:center;align-items:center;transform:translateY(var(--size-6));z-index:var(--layer-2);padding:var(--size-1) var(--size-2);font-size:var(--text-sm);font-family:var(--font-mono);text-align:center}.error.svelte-17v219f.svelte-17v219f{box-shadow:var(--shadow-drop);border:solid 1px var(--error-border-color);border-radius:var(--radius-full);background:var(--error-background-fill);padding-right:var(--size-4);padding-left:var(--size-4);color:var(--error-text-color);font-weight:var(--weight-semibold);font-size:var(--text-lg);line-height:var(--line-lg);font-family:var(--font)}.minimal.svelte-17v219f.svelte-17v219f{pointer-events:none}.minimal.svelte-17v219f .progress-text.svelte-17v219f{background:var(--block-background-fill)}.border.svelte-17v219f.svelte-17v219f{border:1px solid var(--border-color-primary)}.clear-status.svelte-17v219f.svelte-17v219f{position:absolute;display:flex;top:var(--size-2);right:var(--size-2);justify-content:flex-end;gap:var(--spacing-sm);z-index:var(--layer-1)}.toast-body.svelte-syezpc{display:flex;position:relative;right:0;left:0;align-items:center;margin:var(--size-6) var(--size-4);margin:auto;border-radius:var(--container-radius);overflow:hidden;pointer-events:auto}.toast-body.error.svelte-syezpc{border:1px solid var(--color-red-700);background:var(--color-red-50)}.dark .toast-body.error.svelte-syezpc{border:1px solid var(--color-red-500);background-color:var(--color-grey-950)}.toast-body.warning.svelte-syezpc{border:1px solid var(--color-yellow-700);background:var(--color-yellow-50)}.dark .toast-body.warning.svelte-syezpc{border:1px solid var(--color-yellow-500);background-color:var(--color-grey-950)}.toast-body.info.svelte-syezpc{border:1px solid var(--color-grey-700);background:var(--color-grey-50)}.dark .toast-body.info.svelte-syezpc{border:1px solid var(--color-grey-500);background-color:var(--color-grey-950)}.toast-body.success.svelte-syezpc{border:1px solid var(--color-green-700);background:var(--color-green-50)}.dark .toast-body.success.svelte-syezpc{border:1px solid var(--color-green-500);background-color:var(--color-grey-950)}.toast-title.svelte-syezpc{display:flex;align-items:center;font-weight:var(--weight-bold);font-size:var(--text-lg);line-height:var(--line-sm)}.toast-title.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-title.error.svelte-syezpc{color:var(--color-red-50)}.toast-title.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-title.warning.svelte-syezpc{color:var(--color-yellow-50)}.toast-title.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-title.info.svelte-syezpc{color:var(--color-grey-50)}.toast-title.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-title.success.svelte-syezpc{color:var(--color-green-50)}.toast-close.svelte-syezpc{margin:0 var(--size-3);border-radius:var(--size-3);padding:0px var(--size-1-5);font-size:var(--size-5);line-height:var(--size-5)}.toast-close.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-close.error.svelte-syezpc{color:var(--color-red-500)}.toast-close.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-close.warning.svelte-syezpc{color:var(--color-yellow-500)}.toast-close.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-close.info.svelte-syezpc{color:var(--color-grey-500)}.toast-close.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-close.success.svelte-syezpc{color:var(--color-green-500)}.toast-text.svelte-syezpc{font-size:var(--text-lg);word-wrap:break-word;overflow-wrap:break-word;word-break:break-word}.toast-text.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-text.error.svelte-syezpc{color:var(--color-red-50)}.toast-text.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-text.warning.svelte-syezpc{color:var(--color-yellow-50)}.toast-text.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-text.info.svelte-syezpc{color:var(--color-grey-50)}.toast-text.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-text.success.svelte-syezpc{color:var(--color-green-50)}.toast-details.svelte-syezpc{margin:var(--size-3) var(--size-3) var(--size-3) 0;width:100%}.toast-icon.svelte-syezpc{display:flex;position:absolute;position:relative;flex-shrink:0;justify-content:center;align-items:center;margin:var(--size-2);border-radius:var(--radius-full);padding:var(--size-1);padding-left:calc(var(--size-1) - 1px);width:35px;height:35px}.toast-icon.error.svelte-syezpc{color:var(--color-red-700)}.dark .toast-icon.error.svelte-syezpc{color:var(--color-red-500)}.toast-icon.warning.svelte-syezpc{color:var(--color-yellow-700)}.dark .toast-icon.warning.svelte-syezpc{color:var(--color-yellow-500)}.toast-icon.info.svelte-syezpc{color:var(--color-grey-700)}.dark .toast-icon.info.svelte-syezpc{color:var(--color-grey-500)}.toast-icon.success.svelte-syezpc{color:var(--color-green-700)}.dark .toast-icon.success.svelte-syezpc{color:var(--color-green-500)}@keyframes svelte-syezpc-countdown{0%{transform:scaleX(1)}to{transform:scaleX(0)}}.timer.svelte-syezpc{position:absolute;bottom:0;left:0;transform-origin:0 0;animation:svelte-syezpc-countdown 10s linear forwards;width:100%;height:var(--size-1)}.timer.error.svelte-syezpc{background:var(--color-red-700)}.dark .timer.error.svelte-syezpc{background:var(--color-red-500)}.timer.warning.svelte-syezpc{background:var(--color-yellow-700)}.dark .timer.warning.svelte-syezpc{background:var(--color-yellow-500)}.timer.info.svelte-syezpc{background:var(--color-grey-700)}.dark .timer.info.svelte-syezpc{background:var(--color-grey-500)}.timer.success.svelte-syezpc{background:var(--color-green-700)}.dark .timer.success.svelte-syezpc{background:var(--color-green-500)}.hidden.svelte-syezpc{display:none}.toast-text.svelte-syezpc a{text-decoration:underline}.toast-wrap.svelte-gatr8h{display:flex;position:fixed;top:var(--size-4);right:var(--size-4);flex-direction:column;align-items:end;gap:var(--size-2);z-index:var(--layer-top);width:calc(100% - var(--size-8))}@media (--screen-sm){.toast-wrap.svelte-gatr8h{width:calc(var(--size-96) + var(--size-10))}}.streaming-bar.svelte-ga0jj6{position:absolute;bottom:0;left:0;right:0;height:4px;background-color:var(--primary-600);animation:svelte-ga0jj6-countdown linear forwards;z-index:1}@keyframes svelte-ga0jj6-countdown{0%{transform:translate(0)}to{transform:translate(-100%)}}.unstyled-link.svelte-151nsdd{all:unset;cursor:pointer}.panel-container.svelte-1gf2ex4{display:flex;flex-direction:column;border:1px solid var(--border-color-primary);border-radius:0!important;background-color:var(--background-fill-primary);overflow:hidden}.log-view-container.svelte-1gf2ex4{display:flex;flex-direction:column;flex-grow:1;min-height:0}.header.svelte-1gf2ex4{border-bottom:1px solid var(--border-color-primary);background-color:var(--background-fill-secondary);display:flex;justify-content:flex-end;flex-shrink:0}.log-panel.svelte-1gf2ex4{flex-grow:1;font-family:var(--font-mono, monospace);font-size:var(--text-sm);overflow-y:auto;color:#f8f8f8}.log-line.svelte-1gf2ex4{display:flex}.line-number.svelte-1gf2ex4{opacity:.6;padding-right:var(--spacing-lg);-webkit-user-select:none;user-select:none;text-align:right;min-width:3ch}.log-content.svelte-1gf2ex4{margin:0;padding-left:5px;white-space:pre-wrap;word-break:break-word}.log-level-info.svelte-1gf2ex4{color:inherit}.log-level-debug.svelte-1gf2ex4{color:#888}.log-level-warning.svelte-1gf2ex4{color:#facc15}.log-level-error.svelte-1gf2ex4{color:#ef4444}.log-level-critical.svelte-1gf2ex4{background-color:#ef4444;color:#fff;font-weight:700;padding:0 .25rem}.log-level-success.svelte-1gf2ex4{color:#22c55e}.progress-container.svelte-1gf2ex4{padding:var(--spacing-sm) var(--spacing-md);border-top:1px solid var(--border-color-primary);background:var(--background-fill-secondary)}.progress-label-top.svelte-1gf2ex4,.progress-label-bottom.svelte-1gf2ex4{display:flex;justify-content:space-between;font-size:var(--text-sm);color:var(--body-text-color-subdued)}.progress-label-top.svelte-1gf2ex4{margin-bottom:var(--spacing-xs)}.progress-label-bottom.svelte-1gf2ex4{margin-top:var(--spacing-xs)}.progress-bar-background.svelte-1gf2ex4{width:100%;height:8px;background-color:var(--background-fill-primary);border-radius:var(--radius-full);overflow:hidden}.progress-bar-fill.svelte-1gf2ex4{height:100%;background-color:var(--color-accent);border-radius:var(--radius-full);transition:width .1s linear,background-color .3s ease}.progress-bar-fill.success.svelte-1gf2ex4{background-color:var(--color-success, #22c55e)}.progress-bar-fill.error.svelte-1gf2ex4{background-color:var(--color-error, #ef4444)}.rate-info.svelte-1gf2ex4{display:flex;align-items:center;gap:.5ch}.extra-info.svelte-1gf2ex4{color:var(--body-text-color-subdued);font-size:.9em;white-space:nowrap;overflow:hidden;text-overflow:ellipsis;max-width:200px}.block-label-wrapper.svelte-10ojysx{padding-bottom:24px}

src/backend/gradio_livelog/utils.py

@@ -1,10 +1,12 @@
 # backend/gradio_livelog/utils.py
 
+import io
 import logging
 import queue
+import re
 import time
 from contextlib import contextmanager
-from typing import Callable, List, Iterator, Dict, Any
+from typing import Callable, List, Iterator, Dict, Any, Literal, Optional, Union
 
 class _QueueLogHandler(logging.Handler):
     """A private logging handler that directs log records into a queue."""
@@ -16,119 +18,359 @@ class _QueueLogHandler(logging.Handler):
         self.log_queue.put(record)
 
 @contextmanager
-def capture_logs(disable_console: bool = False) -> Iterator[Callable[[], List[logging.LogRecord]]]:
+def capture_logs(
+    log_name: Union[str, List[str], None] = None, 
+    log_level: int = logging.INFO,
+    disable_console: bool = False
+) -> Iterator[Callable[[], List[logging.LogRecord]]]:
     """
-    A context manager to capture logs from the root logger.
+    A context manager to capture logs from one or more specified loggers.
 
-    Temporarily attaches a handler to the root logger to intercept all log
-    messages. If `disable_console` is True, it will also temporarily remove
-    other console-based StreamHandlers to prevent duplicate output.
+    This function temporarily attaches a thread-safe, queue-based handler to the
+    target logger(s) to intercept all log messages. If `disable_console` is True,
+    it will also temporarily remove other console-based StreamHandlers from the
+    target loggers to prevent duplicate output to the terminal.
 
     Args:
-        disable_console: If True, prevents logs from also being printed to the console.
+        log_name: The name of the logger(s) to capture.
+                - `str`: Captures logs from a single named logger.
+                - `List[str]`: Captures logs from multiple named loggers.
+                - `None` or `""`: Captures logs from the root logger.
+        log_level: The minimum level of logs to capture (e.g., `logging.INFO`).
+        disable_console: If True, prevents the captured logs from also being
+                        printed to the console by other handlers on the same logger.
 
     Yields:
-        A function that, when called, returns a list of all log records captured
-        since the last call.
+        A callable function. When this function is called, it returns a list
+        of all log records captured since the last time it was called, effectively
+        acting as a "get new logs" utility.
+        
+    Example:
+        >>> with capture_logs(log_name=["my_app", "my_library"]) as get_logs:
+        ...     logging.getLogger("my_app").info("Starting process.")
+        ...     new_logs = get_logs()  # Contains the first log record
+        ...     logging.getLogger("my_library").warning("A potential issue.")
+        ...     more_logs = get_logs() # Contains only the warning record
     """
+    # Step 1: Determine the target loggers based on the `log_name` argument.
+    target_loggers: List[logging.Logger] = []
+    log_names_to_process = []
+    if log_name is None or log_name == "":
+        log_names_to_process.append(None) # `None` is the identifier for the root logger
+    elif isinstance(log_name, list):
+        log_names_to_process.extend(log_name)
+    elif isinstance(log_name, str):
+        log_names_to_process.append(log_name)
+    
+    # Get the actual logger objects from their names.
+    for name in set(log_names_to_process): # Use set to avoid duplicates
+        target_loggers.append(logging.getLogger(name))
+
+    # Step 2: Set up the thread-safe queue and the custom handler.
     log_queue = queue.Queue()
     queue_handler = _QueueLogHandler(log_queue)
-    root_logger = logging.getLogger()
 
-    original_level = root_logger.level
-    root_logger.setLevel(logging.DEBUG)
-    root_logger.addHandler(queue_handler)
+    # Step 3: Store the original state of each logger to restore it later.
+    original_levels = {logger.name: logger.level for logger in target_loggers}
+    original_handlers = {logger.name: logger.handlers[:] for logger in target_loggers}
+    
+    # Step 4: Modify the target loggers for the duration of the context.
+    for logger in target_loggers:
+        # Set the desired capture level.
+        logger.setLevel(log_level)
 
-    removed_handlers = []
-    if disable_console:
-        for handler in root_logger.handlers[:]: # Iterate over a copy
-            if isinstance(handler, logging.StreamHandler) and handler is not queue_handler:
-                removed_handlers.append(handler)
-                root_logger.removeHandler(handler)
+        if disable_console:
+            # If disabling console, remove all existing StreamHandlers.
+            # We keep other handlers (e.g., FileHandler) intact.
+            logger.handlers = [
+                h for h in logger.handlers if not isinstance(h, logging.StreamHandler)
+            ]
+        
+        # Add our custom queue handler to start capturing logs.
+        logger.addHandler(queue_handler)
 
-    all_captured = [] 
-    last_returned = 0 
+    # This holds all records captured during the context's lifetime.
+    all_captured: List[logging.LogRecord] = [] 
+    # This index tracks the last record that was returned to the caller.
+    last_returned_index = 0 
 
     try:
-        def get_captured_records():
-            nonlocal last_returned
-            while True:
+        def get_captured_records() -> List[logging.LogRecord]:
+            """
+            Retrieves new log records from the queue and returns them.
+            This function is what the context manager yields to the user.
+            """
+            nonlocal last_returned_index
+            
+            # Drain the queue into our master list of captured records.
+            while not log_queue.empty():
                 try:
-                    all_captured.append(log_queue.get_nowait()) 
+                    record = log_queue.get_nowait()
+                    all_captured.append(record)
                 except queue.Empty:
-                    break
-            new_records = all_captured[last_returned:]
-            last_returned = len(all_captured)
+                    # This handles a rare race condition where the queue becomes empty
+                    # between the `empty()` check and `get_nowait()`.
+                    break 
+            
+            # Slice the master list to get only the new records.
+            new_records = all_captured[last_returned_index:]
+            # Update the index to the end of the list for the next call.
+            last_returned_index = len(all_captured)
+            
             return new_records
+
+        # Yield the function to the `with` block.
         yield get_captured_records
+
     finally:
-        root_logger.removeHandler(queue_handler)
-        root_logger.setLevel(original_level)
-        for handler in removed_handlers:
-            root_logger.addHandler(handler)
+        # Step 5: Restore the loggers to their original state, ensuring no side effects.
+        for logger in target_loggers:
+            # Remove our custom handler.
+            logger.removeHandler(queue_handler)
+            
+            # Restore the original log level.
+            if logger.name in original_levels:
+                logger.setLevel(original_levels[logger.name])
+            
+            # If we disabled the console, restore the original handlers.
+            if disable_console and logger.name in original_handlers:
+                # It's safest to clear handlers and then re-add the originals.
+                logger.handlers = []
+                for handler in original_handlers[logger.name]:
+                    logger.addHandler(handler)
+                    
+class Tee(io.StringIO):
+    """
+    A file-like object that acts like the Unix 'tee' command.
+    It writes to multiple file-like objects simultaneously.
+    """
+    def __init__(self, *files):
+        """
+        Initializes the Tee object.
+        Args:
+            *files: A variable number of file-like objects (e.g., sys.stderr,
+                    a TqdmToQueueWriter instance, etc.).
+        """
+        super().__init__()
+        self.files = files
+
+    def write(self, s: str) -> int:
+        """
+        Writes the string 's' to all managed files.
+        """
+        for f in self.files:
+            f.write(s)
+            # Some file-like objects, like the console, might need to be flushed.
+            if hasattr(f, 'flush'):
+                f.flush()
+        return len(s)
+
+    def flush(self):
+        """Flushes all managed files."""
+        for f in self.files:
+            if hasattr(f, 'flush'):
+                f.flush()
+class TqdmToQueueWriter(io.StringIO):
+    """
+    A custom, thread-safe, file-like object that intercepts tqdm's output.
+
+    This class is designed to be passed to a `tqdm` instance (or a library
+    that uses `tqdm`, like `diffusers`) via its `file` argument. It uses a
+    regular expression to parse the formatted progress string in real-time.
+
+    It extracts key metrics:
+    - The iteration rate value (e.g., 2.73).
+    - The rate unit ("it/s" or "s/it").
+    - Any additional status information that follows the rate.
+
+    The extracted data is packaged into a dictionary and put onto a
+    `queue.Queue`, allowing a consumer thread (like a Gradio UI thread)
+    to receive real-time progress data from a worker thread.
+    """
+    def __init__(self, rate_queue: queue.Queue):
+        """
+        Initializes the writer with a queue for communication.
+
+        Args:
+            rate_queue (queue.Queue): The thread-safe queue to which the
+                                      extracted rate data will be sent.
+        """
+        super().__init__()
+        self.rate_queue = rate_queue
+        # Regex Explanation:
+        # (\d+\.?\d*): Group 1, captures the rate value (float or integer).
+        # \s*: Matches any whitespace.
+        # (it/s|s/it): Group 2, captures the rate unit, which can be "it/s" or "s/it".
+        # ,\s*: Matches the comma and whitespace separator.
+        # (.*): Group 3, greedily captures all remaining characters in the line.
+        self.rate_regex = re.compile(r"(\d+\.?\d*)\s*(it/s|s/it)(?:,\s*(.*))?")
+
+    def write(self, s: str) -> int:
+        """
+        This method is called by `tqdm` whenever it updates the progress bar.
+        It receives the full, formatted progress string.
+
+        Args:
+            s (str): The string output from `tqdm` (e.g., "75%|...| 2.73it/s, ...").
+
+        Returns:
+            int: The number of characters written, as required by the file-like
+                 object interface.
+        """
+        match = self.rate_regex.search(s)
+        if match:
+            try:
+                rate_value = float(match.group(1))
+                rate_unit = match.group(2)                
+                extra_info_raw = match.group(3) 
+
+                cleaned_info = ""
+                if extra_info_raw:                    
+                    cleaned_info = extra_info_raw.rstrip(']').strip()
+                    cleaned_info = cleaned_info.replace('[', ' ').replace(']', ' ')
+                    cleaned_info = re.sub(r'\s+', ' ', cleaned_info).strip()
 
+                # Put a structured dictionary onto the queue.
+                self.rate_queue.put({
+                    "rate": rate_value,
+                    "unit": rate_unit,
+                    "extra": cleaned_info
+                })
+            except (ValueError, IndexError):
+                # Silently ignore parsing errors if the regex match is malformed.
+                pass
+        
+        # Fulfill the file-like object contract.
+        return len(s)
+                    
 class ProgressTracker:
     """
     A helper class to track and format progress updates for the LiveLog component.
 
-    This class mimics some of the behavior of `tqdm`, calculating the rate of
-    iterations per second and providing a structured dictionary for easy use
-    with Gradio's `yield` mechanism.
+    This versatile class operates in a hybrid mode for calculating iteration rates:
+    1.  **Internal Calculation (Default):** It uses an Exponential Moving Average (EMA)
+        to compute a smoothed, stable rate. The unit for this internal calculation
+        (`it/s` or `s/it`) can be specified during initialization, making it flexible
+        for different types of processes.
+    2.  **External Override (Preferred):** It can accept a dictionary of externally
+        captured rate data (e.g., from a `tqdm` instance). This provides the most
+        accurate possible display by sourcing the rate and its unit directly from
+        the process being monitored, overriding any internal calculations.
+
+    The tracker also intelligently "freezes" the last known rate when the process
+    status changes to 'success' or 'error', ensuring the final speed remains visible on the UI.
     """
-    def __init__(self, total: int, description: str = "Processing..."):
+    def __init__(self, total: int, description: str = "Processing...", 
+                 smoothing_factor: float = 0.3, 
+                 rate_unit: Literal["it/s", "s/it"] = "s/it"):
         """
         Initializes the progress tracker.
 
         Args:
-            total: The total number of iterations for the process.
-            description: A short, fixed description of the task being performed.
+            total (int): The total number of iterations for the process.
+            description (str): A short, fixed description of the task being performed.
+            smoothing_factor (float): The EMA smoothing factor used for the internal
+                                      rate calculation. A smaller value (e.g., 0.1)
+                                      results in smoother but less responsive updates.
+            rate_unit (Literal["it/s", "s/it"]): The preferred unit for the
+                                                 internal rate calculation when no
+                                                 external data is provided. Defaults to "it/s".
         """
         self.total = total
         self.description = description
+        self.smoothing_factor = smoothing_factor
+        self.preferred_rate_unit = rate_unit  # Stores the user's preference for internal calculations.
+        
         self.current = 0
         self.start_time = time.time()
         self.last_update_time = self.start_time
         self.last_update_item = 0
+        
+        # State fields that will be updated and returned.
         self.rate = 0.0
+        self.rate_unit = self.preferred_rate_unit  # Sets the initial unit.
+        self.extra_info = ""
 
-    def update(self, advance: int = 1, status: str = "running", logs: List[Dict] = None, log_content: str = None) -> Dict[str, Any]:
+    def update(self, advance: int = 1, status: str = "running", 
+               logs: Optional[List[Dict]] = None, log_content: Optional[str] = None, 
+               rate_data: Optional[Dict] = None) -> Dict[str, Any]:
         """
         Advances the progress and returns a dictionary formatted for the LiveLog component.
 
+        This method forms the core of the tracker, updating the current progress
+        and determining the iteration rate based on the provided parameters.
+
         Args:
-            advance: The number of steps to advance the progress by (default is 1).
-            status: The current status of the process ("running", "success", "error").
-            logs: An optional list of all log dictionaries generated so far. If provided,
-                  this list will be passed to the frontend, allowing the log view to be
-                  updated simultaneously with the progress bar.
-            log_content: An optional string to use as the progress bar's description for this
-                         specific update, overriding the fixed description. This is useful
-                         for showing the most recent log message as the progress description.
+            advance (int): The number of steps to advance the progress by (default is 1).
+            status (str): The current status of the process ("running", "success", "error").
+            logs (Optional[List[Dict]]): An optional list of log dictionaries to pass to the frontend.
+            log_content (Optional[str]): An optional string to override the fixed description for this update.
+            rate_data (Optional[Dict]): A dictionary from an external source (like `tqdm`)
+                                        containing keys like 'rate', 'unit', and 'extra'.
+                                        If provided, this data will override all internal
+                                        rate calculations.
 
         Returns:
-            A dictionary formatted for the LiveLog component's frontend.
+            Dict[str, Any]: A state dictionary formatted for the LiveLog component's frontend.
         """
         self.current += advance
         self.current = min(self.current, self.total)
 
         now = time.time()
-        delta_time = now - self.last_update_time
-        delta_items = self.current - self.last_update_item
+        
+        # --- Enhanced Hybrid Rate Logic ---
+        
+        # Priority 1: Use the externally provided rate data if it exists. This is the most accurate source.
+        if rate_data:
+            self.rate = rate_data.get("rate", self.rate)
+            self.rate_unit = rate_data.get("unit", self.rate_unit)
+            self.extra_info = rate_data.get("extra", self.extra_info)
+        
+        # Priority 2: If no external data, perform internal calculation, but only if the
+        # process is actively running and has advanced.
+        elif status == "running" and advance > 0:
+            delta_time = now - self.last_update_time
+            delta_items = self.current - self.last_update_item
+
+            # Only calculate if time has passed and items have progressed to avoid division by zero.
+            if delta_time > 0 and delta_items > 0:
+                
+                # Calculate the rate based on the user's preferred unit.
+                if self.preferred_rate_unit == "it/s":
+                    instant_rate = delta_items / delta_time
+                    self.rate_unit = "it/s"
+                else:  # "s/it"
+                    instant_rate = delta_time / delta_items
+                    self.rate_unit = "s/it"
+
+                # Apply Exponential Moving Average (EMA) for a smoother reading.
+                if self.rate == 0.0: # Use the first measurement as the initial value.
+                    self.rate = instant_rate
+                else:
+                    self.rate = (self.smoothing_factor * instant_rate) + \
+                                ((1 - self.smoothing_factor) * self.rate)
+                
+                # Internal calculations do not produce extra info.
+                self.extra_info = ""
 
-        # Stabilize the rate calculation by updating it periodically or at the very end.
-        if delta_time > 0.1 or self.current == self.total:
-            self.rate = delta_items / delta_time if delta_time > 0 else 0.0
             self.last_update_time = now
             self.last_update_item = self.current
         
+        # Priority 3: If status is 'success' or 'error', or if advance is 0,
+        # do nothing to the rate fields. This "freezes" them at their last known values.
+        
+        # Determine the description to display for this specific update.
         desc = log_content if log_content is not None else self.description
         
+        # Assemble and return the final state dictionary for the frontend.
         return {
             "type": "progress",
             "current": self.current,
             "total": self.total,
             "desc": desc,
             "rate": self.rate,
+            "rate_unit": self.rate_unit,
+            "extra_info": self.extra_info,
             "status": status,
             "logs": logs or [],
         }

src/demo/app.py

@@ -1,8 +1,8 @@
 # demo/app.py
 
+import sys
 import gradio as gr
 import torch
-import time
 import logging
 import random
 import numpy as np
@@ -14,24 +14,28 @@ import spaces
 
 # Import the component and ALL its utilities
 from gradio_livelog import LiveLog
-from gradio_livelog.utils import ProgressTracker, capture_logs
+from gradio_livelog.utils import ProgressTracker, Tee, TqdmToQueueWriter, capture_logs
 
 # --- 1. SETUP ---
 MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
 MAX_SEED = np.iinfo(np.int32).max
 
 # --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
-logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
+app_logger = logging.getLogger("logging_app")
+app_logger.setLevel(logging.INFO)
+console_handler = logging.StreamHandler()
+console_handler.flush = sys.stderr.flush
+app_logger.addHandler(console_handler)
 
-async def run_process(disable_console: bool, run_error_case: bool):
-    with capture_logs(disable_console=disable_console) as get_logs:
+async def run_process(disable_console: bool, rate_unit: str, run_error_case: bool):
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
         total_steps = 100
-        tracker = ProgressTracker(total=total_steps, description="Simulating a process...")
+        tracker = ProgressTracker(total=total_steps, description="Simulating a process...", rate_unit=rate_unit)
         all_logs = []
         last_log_content = None
         
         initial_log = f"Starting simulated process with {total_steps} steps..."
-        logging.info(initial_log)
+        app_logger.info(initial_log)
         logs = [
             {
                 "type": "log",
@@ -49,14 +53,14 @@ async def run_process(disable_console: bool, run_error_case: bool):
             current_step = i + 1
             
             if current_step == 10:
-                logging.warning(f"Low disk space warning at step {current_step}.")
+                app_logger.warning(f"Low disk space warning at step {current_step}.")
             elif current_step == 30:
-                logging.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
+                app_logger.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
             elif current_step == 75:
-                logging.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
+                app_logger.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
             
             if run_error_case and current_step == 50:
-                logging.error("A fatal simulation error occurred! Aborting.")
+                app_logger.error("A fatal simulation error occurred! Aborting.")
                 logs = [
                     {
                         "type": "log",
@@ -84,7 +88,7 @@ async def run_process(disable_console: bool, run_error_case: bool):
             yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
         
         final_log = "Process completed successfully!"
-        logging.log(logging.INFO + 5, final_log)
+        app_logger.log(logging.INFO + 5, final_log)
         logs = [
             {
                 "type": "log",
@@ -103,47 +107,60 @@ def update_livelog_properties(mode, color, lines, scroll):
 def clear_output():
     return None
 
-async def run_success_case(disable_console: bool):
+async def run_success_case(disable_console: bool, rate_unit: str):
     yield None    
-    async for update in run_process(disable_console=disable_console, run_error_case=False):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=False):
         yield update
 
-async def run_error_case(disable_console: bool):
+async def run_error_case(disable_console: bool, rate_unit: str):
     yield None
-    async for update in run_process(disable_console=disable_console, run_error_case=True):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=True):
         yield update
 
 
 # --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
 diffusion_pipeline = None
+pipeline_lock = threading.Lock()
 def load_pipeline(on_load=True):
     """A function to load the model, ensuring it's only done once."""
     global diffusion_pipeline
-    if diffusion_pipeline is None:
-        print("Loading Stable Diffusion model for the first time...")
-        pipe = StableDiffusionXLPipeline.from_pretrained(
-            MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
-        )
-        pipe.enable_vae_tiling()
-        pipe.enable_model_cpu_offload()
-        pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-        pipe.set_progress_bar_config(disable=True)
-        diffusion_pipeline = pipe
-        print("Model loaded successfully.")
-    
+    with pipeline_lock:
+        if diffusion_pipeline is None:
+            print("Loading Stable Diffusion model for the first time...")
+            pipe = StableDiffusionXLPipeline.from_pretrained(
+                MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
+            )
+            pipe.enable_vae_tiling()
+            pipe.enable_model_cpu_offload()
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)        
+            diffusion_pipeline = pipe
+            print("Model loaded successfully!")
+        
     if not on_load:
         return diffusion_pipeline
 
 @spaces.GPU(duration=60, enable_queue=True)
-def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
+def run_diffusion_in_thread(prompt: str, disable_console: bool, update_queue: queue.Queue):
     """
     This function now uses capture_logs to listen to internal diffusers logs
     while retaining the superior data structure you designed.
     """
     tracker = None    
-    with capture_logs() as get_logs:
-        try:
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        try:            
             pipe = load_pipeline(on_load=False)
+            
+            #We will capture pipeline tqdm s/it progress instead            
+            rate_queue = queue.Queue()
+            tqdm_writer = TqdmToQueueWriter(rate_queue)
+            
+            progress_bar_handler = Tee(sys.stderr, tqdm_writer)
+            pipe.set_progress_bar_config(file=progress_bar_handler,  #if you dont need to see the tqdm progress on console set file=tqdm_writer instead            
+                                        disable=False,  
+                                        ncols=100,
+                                        dynamic_ncols=True,
+                                        ascii=" █")
+            
             seed = random.randint(0, MAX_SEED)
             generator = torch.Generator(device="cuda").manual_seed(seed)
             prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
@@ -151,67 +168,110 @@ def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
             num_inference_steps = 10
             
             all_logs = []
-            last_log_content = None
+            last_known_rate_data = None
 
             # Helper function to process and store new logs
-            def process_and_store_logs():
-                nonlocal all_logs, last_log_content
+            def process_and_send_updates(status="running", advance=0, final_image_payload=None):
+                """
+                This is the core callback function. It captures new logs, formats them,
+                and sends a complete update object (logs + progress) to the UI queue.
+                This should also be called after the log record.
+                """
+                nonlocal all_logs, last_known_rate_data
+                new_rate_data = None
+                while not rate_queue.empty():
+                    try:
+                        new_rate_data = rate_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                
+                if new_rate_data is not None:
+                    last_known_rate_data = new_rate_data
+                
                 new_records = get_logs()
                 if new_records:
-                    new_logs = [{"type": "log", "level": r.levelname, "content": r.getMessage()} for r in new_records]
+                    new_logs = [{
+                        "type": "log",
+                        "level": "SUCCESS" if r.levelno == logging.INFO + 5 else r.levelname,
+                        "content": r.getMessage()
+                    } for r in new_records]
                     all_logs.extend(new_logs)
-                    last_log_content = all_logs[-1]["content"]
-            
-            logging.info(f"Using seed: {seed}")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            logging.info("Starting diffusion process...")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps")
+                
+                # Use the tracker to generate the progress update dictionary if it exists.
+                # If not, create a preliminary update dictionary.
+                update_dict = {}
+                
+                if tracker:
+                    update_dict = tracker.update(
+                        advance=advance, 
+                        status=status, 
+                        logs=all_logs,
+                        rate_data=last_known_rate_data                               
+                    )
+                else:
+                    # Initial state before the tracker is created.
+                    update_dict = {
+                        "type": "progress", 
+                        "logs": all_logs, 
+                        "current": 0, 
+                        "total": num_inference_steps, 
+                        "desc": "Diffusion Steps" # Description is sent once
+                    }
+
+                # Put the update on the queue. The image payload is usually None
+                # until the very end.
+                update_queue.put((final_image_payload, update_dict))
+                
+            app_logger.info(f"Using seed: {seed}")
+            process_and_send_updates()
+                        
+            app_logger.info("Starting diffusion process...")
+            process_and_send_updates()
+                        
+            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps", rate_unit='it/s')
             
             def progress_callback(pipe_instance, step, timestep, callback_kwargs):
-                process_and_store_logs() # Check for new logs from diffusers at each step
-                update_dict = tracker.update(logs=all_logs)
-                update_queue.put((None, update_dict))
+                process_and_send_updates(advance=1) 
                 return callback_kwargs
-            
+                        
             images = pipe(
                 prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
                 guidance_scale=3.0, num_inference_steps=num_inference_steps,
                 generator=generator, callback_on_step_end=progress_callback
             ).images
             
-            logging.log(logging.INFO + 5, "Image generated successfully!")
-            process_and_store_logs()
-            
-            final_update = tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-            update_queue.put((images, final_update))
+            app_logger.log(logging.INFO + 5, "Image generated successfully!")
+            process_and_send_updates(status="success", final_image_payload=images)
+           
 
         except Exception as e:
-            logging.error(f"Error in diffusion thread: {e}", exc_info=True)
-            process_and_store_logs() # Capture the final error log
-            if tracker:
-                error_update = tracker.update(advance=0, status="error", logs=all_logs, log_content=f"An error occurred: {e}")
-                update_queue.put((None, error_update))
+            app_logger.error(f"Error in diffusion thread: {e}, process aborted!", exc_info=True)                    
+            process_and_send_updates(status="error")                                                        
         finally:
             update_queue.put(None)
             
+            
 @spaces.GPU(duration=60, enable_queue=True)
 def generate(prompt):
     """This function starts the worker thread and yields updates from the queue."""   
-    yield None, None
-    yield None, {"type": "log", "level": "INFO", "content": "Preparing generation..."}
+    yield None, None, gr.update(interactive=False)    
     update_queue = queue.Queue()
-    diffusion_thread = threading.Thread(target=run_diffusion_in_thread, args=(prompt, update_queue))
+    diffusion_thread = threading.Thread(target=run_diffusion_in_thread,  args=(prompt, False, update_queue))
     diffusion_thread.start()
+    final_images = None
     while True:
         update = update_queue.get()
-        if update is None: break
-        yield update
-
+        if update is None: 
+            break
+        
+        images, log_update = update
+        
+        if images:
+            final_images = images
+      
+        yield final_images, log_update, gr.skip()
+    
+    yield final_images, log_update, gr.update(interactive=True)
 
 # --- 4. THE COMBINED GRADIO UI with TABS ---
 with gr.Blocks(theme=gr.themes.Ocean()) as demo:
@@ -230,17 +290,18 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                     with gr.Group():
                         gr.Markdown("### Component Properties")
                         display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
+                        rate_unit = gr.Radio(["it/s","s/it"], label="Progress rate unit", value="it/s")
                         bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
                         line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
                         autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
+                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                     with gr.Group():
                         gr.Markdown("### Simulation Controls")
-                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                         start_btn = gr.Button("Run Success Case", variant="primary")
                         error_btn = gr.Button("Run Error Case")
             
-            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox], outputs=feature_logger)
+            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
             feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
             controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
             for control in controls:
@@ -267,7 +328,7 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                         label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
                     )
             
-            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
+            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer, run_button])
             prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
             livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
             

src/demo/space.py

@@ -9,7 +9,7 @@ abs_path = os.path.join(os.path.dirname(__file__), "css.css")
 
 with gr.Blocks(
     css=abs_path,
-    theme=gr.themes.Ocean(
+    theme=gr.themes.Default(
         font_mono=[
             gr.themes.GoogleFont("Inconsolata"),
             "monospace",
@@ -40,9 +40,9 @@ pip install gradio_livelog
 ```python
 # demo/app.py
 
+import sys
 import gradio as gr
 import torch
-import time
 import logging
 import random
 import numpy as np
@@ -54,24 +54,28 @@ import spaces
 
 # Import the component and ALL its utilities
 from gradio_livelog import LiveLog
-from gradio_livelog.utils import ProgressTracker, capture_logs
+from gradio_livelog.utils import ProgressTracker, Tee, TqdmToQueueWriter, capture_logs
 
 # --- 1. SETUP ---
 MODEL_ID = "SG161222/RealVisXL_V5.0_Lightning"
 MAX_SEED = np.iinfo(np.int32).max
 
 # --- 2. LOGIC FOR THE "LIVELOG FEATURE DEMO" TAB ---
-logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
-
-async def run_process(disable_console: bool, run_error_case: bool):
-    with capture_logs(disable_console=disable_console) as get_logs:
+app_logger = logging.getLogger("logging_app")
+app_logger.setLevel(logging.INFO)
+console_handler = logging.StreamHandler()
+console_handler.flush = sys.stderr.flush
+app_logger.addHandler(console_handler)
+
+async def run_process(disable_console: bool, rate_unit: str, run_error_case: bool):
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
         total_steps = 100
-        tracker = ProgressTracker(total=total_steps, description="Simulating a process...")
+        tracker = ProgressTracker(total=total_steps, description="Simulating a process...", rate_unit=rate_unit)
         all_logs = []
         last_log_content = None
         
         initial_log = f"Starting simulated process with {total_steps} steps..."
-        logging.info(initial_log)
+        app_logger.info(initial_log)
         logs = [
             {
                 "type": "log",
@@ -89,14 +93,14 @@ async def run_process(disable_console: bool, run_error_case: bool):
             current_step = i + 1
             
             if current_step == 10:
-                logging.warning(f"Low disk space warning at step {current_step}.")
+                app_logger.warning(f"Low disk space warning at step {current_step}.")
             elif current_step == 30:
-                logging.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
+                app_logger.log(logging.INFO + 5, f"Asset pack loaded successfully at step {current_step}.")
             elif current_step == 75:
-                logging.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
+                app_logger.critical(f"Checksum mismatch! Data may be corrupt at step {current_step}.")
             
             if run_error_case and current_step == 50:
-                logging.error("A fatal simulation error occurred! Aborting.")
+                app_logger.error("A fatal simulation error occurred! Aborting.")
                 logs = [
                     {
                         "type": "log",
@@ -124,7 +128,7 @@ async def run_process(disable_console: bool, run_error_case: bool):
             yield tracker.update(advance=1, status="running", logs=all_logs, log_content=last_log_content)
         
         final_log = "Process completed successfully!"
-        logging.log(logging.INFO + 5, final_log)
+        app_logger.log(logging.INFO + 5, final_log)
         logs = [
             {
                 "type": "log",
@@ -143,47 +147,60 @@ def update_livelog_properties(mode, color, lines, scroll):
 def clear_output():
     return None
 
-async def run_success_case(disable_console: bool):
+async def run_success_case(disable_console: bool, rate_unit: str):
     yield None    
-    async for update in run_process(disable_console=disable_console, run_error_case=False):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=False):
         yield update
 
-async def run_error_case(disable_console: bool):
+async def run_error_case(disable_console: bool, rate_unit: str):
     yield None
-    async for update in run_process(disable_console=disable_console, run_error_case=True):
+    async for update in run_process(disable_console=disable_console, rate_unit=rate_unit, run_error_case=True):
         yield update
 
 
 # --- 3. LOGIC FOR THE "DIFFUSION PIPELINE INTEGRATION" TAB ---
 diffusion_pipeline = None
+pipeline_lock = threading.Lock()
 def load_pipeline(on_load=True):
     \"\"\"A function to load the model, ensuring it's only done once.\"\"\"
     global diffusion_pipeline
-    if diffusion_pipeline is None:
-        print("Loading Stable Diffusion model for the first time...")
-        pipe = StableDiffusionXLPipeline.from_pretrained(
-            MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
-        )
-        pipe.enable_vae_tiling()
-        pipe.enable_model_cpu_offload()
-        pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-        pipe.set_progress_bar_config(disable=True)
-        diffusion_pipeline = pipe
-        print("Model loaded successfully.")
-    
+    with pipeline_lock:
+        if diffusion_pipeline is None:
+            print("Loading Stable Diffusion model for the first time...")
+            pipe = StableDiffusionXLPipeline.from_pretrained(
+                MODEL_ID, torch_dtype=torch.float16, use_safetensors=True, add_watermarker=False
+            )
+            pipe.enable_vae_tiling()
+            pipe.enable_model_cpu_offload()
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)        
+            diffusion_pipeline = pipe
+            print("Model loaded successfully!")
+        
     if not on_load:
         return diffusion_pipeline
 
 @spaces.GPU(duration=60, enable_queue=True)
-def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
+def run_diffusion_in_thread(prompt: str, disable_console: bool, update_queue: queue.Queue):
     \"\"\"
     This function now uses capture_logs to listen to internal diffusers logs
     while retaining the superior data structure you designed.
     \"\"\"
     tracker = None    
-    with capture_logs() as get_logs:
-        try:
+    with capture_logs(log_level=logging.INFO, log_name=["logging_app"], disable_console=disable_console) as get_logs: #You can watch more than one log if you wish in log_name. If you do not pass log_name, the default log will be used.
+        try:            
             pipe = load_pipeline(on_load=False)
+            
+            #We will capture pipeline tqdm s/it progress instead            
+            rate_queue = queue.Queue()
+            tqdm_writer = TqdmToQueueWriter(rate_queue)
+            
+            progress_bar_handler = Tee(sys.stderr, tqdm_writer)
+            pipe.set_progress_bar_config(file=progress_bar_handler,  #if you dont need to see the tqdm progress on console set file=tqdm_writer instead            
+                                        disable=False,  
+                                        ncols=100,
+                                        dynamic_ncols=True,
+                                        ascii=" █")
+            
             seed = random.randint(0, MAX_SEED)
             generator = torch.Generator(device="cuda").manual_seed(seed)
             prompt_style = f"hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic"
@@ -191,67 +208,110 @@ def run_diffusion_in_thread(prompt: str, update_queue: queue.Queue):
             num_inference_steps = 10
             
             all_logs = []
-            last_log_content = None
+            last_known_rate_data = None
 
             # Helper function to process and store new logs
-            def process_and_store_logs():
-                nonlocal all_logs, last_log_content
+            def process_and_send_updates(status="running", advance=0, final_image_payload=None):
+                \"\"\"
+                This is the core callback function. It captures new logs, formats them,
+                and sends a complete update object (logs + progress) to the UI queue.
+                This should also be called after the log record.
+                \"\"\"
+                nonlocal all_logs, last_known_rate_data
+                new_rate_data = None
+                while not rate_queue.empty():
+                    try:
+                        new_rate_data = rate_queue.get_nowait()
+                    except queue.Empty:
+                        break
+                
+                if new_rate_data is not None:
+                    last_known_rate_data = new_rate_data
+                
                 new_records = get_logs()
                 if new_records:
-                    new_logs = [{"type": "log", "level": r.levelname, "content": r.getMessage()} for r in new_records]
+                    new_logs = [{
+                        "type": "log",
+                        "level": "SUCCESS" if r.levelno == logging.INFO + 5 else r.levelname,
+                        "content": r.getMessage()
+                    } for r in new_records]
                     all_logs.extend(new_logs)
-                    last_log_content = all_logs[-1]["content"]
-            
-            logging.info(f"Using seed: {seed}")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            logging.info("Starting diffusion process...")
-            process_and_store_logs()
-            update_queue.put((None, {"type": "progress", "logs": all_logs, "current": 0, "total": num_inference_steps, "desc": "Diffusion Steps"}))
-            
-            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps")
+                
+                # Use the tracker to generate the progress update dictionary if it exists.
+                # If not, create a preliminary update dictionary.
+                update_dict = {}
+                
+                if tracker:
+                    update_dict = tracker.update(
+                        advance=advance, 
+                        status=status, 
+                        logs=all_logs,
+                        rate_data=last_known_rate_data                               
+                    )
+                else:
+                    # Initial state before the tracker is created.
+                    update_dict = {
+                        "type": "progress", 
+                        "logs": all_logs, 
+                        "current": 0, 
+                        "total": num_inference_steps, 
+                        "desc": "Diffusion Steps" # Description is sent once
+                    }
+
+                # Put the update on the queue. The image payload is usually None
+                # until the very end.
+                update_queue.put((final_image_payload, update_dict))
+                
+            app_logger.info(f"Using seed: {seed}")
+            process_and_send_updates()
+                        
+            app_logger.info("Starting diffusion process...")
+            process_and_send_updates()
+                        
+            tracker = ProgressTracker(total=num_inference_steps, description="Diffusion Steps", rate_unit='it/s')
             
             def progress_callback(pipe_instance, step, timestep, callback_kwargs):
-                process_and_store_logs() # Check for new logs from diffusers at each step
-                update_dict = tracker.update(logs=all_logs)
-                update_queue.put((None, update_dict))
+                process_and_send_updates(advance=1) 
                 return callback_kwargs
-            
+                        
             images = pipe(
                 prompt=prompt_style, negative_prompt=negative_prompt_style, width=1024, height=1024,
                 guidance_scale=3.0, num_inference_steps=num_inference_steps,
                 generator=generator, callback_on_step_end=progress_callback
             ).images
             
-            logging.log(logging.INFO + 5, "Image generated successfully!")
-            process_and_store_logs()
-            
-            final_update = tracker.update(advance=0, status="success", logs=all_logs, log_content=last_log_content)
-            update_queue.put((images, final_update))
+            app_logger.log(logging.INFO + 5, "Image generated successfully!")
+            process_and_send_updates(status="success", final_image_payload=images)
+           
 
         except Exception as e:
-            logging.error(f"Error in diffusion thread: {e}", exc_info=True)
-            process_and_store_logs() # Capture the final error log
-            if tracker:
-                error_update = tracker.update(advance=0, status="error", logs=all_logs, log_content=f"An error occurred: {e}")
-                update_queue.put((None, error_update))
+            app_logger.error(f"Error in diffusion thread: {e}, process aborted!", exc_info=True)                    
+            process_and_send_updates(status="error")                                                        
         finally:
             update_queue.put(None)
             
+            
 @spaces.GPU(duration=60, enable_queue=True)
 def generate(prompt):
     \"\"\"This function starts the worker thread and yields updates from the queue.\"\"\"   
-    yield None, None
-    yield None, {"type": "log", "level": "INFO", "content": "Preparing generation..."}
+    yield None, None, gr.update(interactive=False)    
     update_queue = queue.Queue()
-    diffusion_thread = threading.Thread(target=run_diffusion_in_thread, args=(prompt, update_queue))
+    diffusion_thread = threading.Thread(target=run_diffusion_in_thread,  args=(prompt, False, update_queue))
     diffusion_thread.start()
+    final_images = None
     while True:
         update = update_queue.get()
-        if update is None: break
-        yield update
-
+        if update is None: 
+            break
+        
+        images, log_update = update
+        
+        if images:
+            final_images = images
+      
+        yield final_images, log_update, gr.skip()
+    
+    yield final_images, log_update, gr.update(interactive=True)
 
 # --- 4. THE COMBINED GRADIO UI with TABS ---
 with gr.Blocks(theme=gr.themes.Ocean()) as demo:
@@ -270,17 +330,18 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                     with gr.Group():
                         gr.Markdown("### Component Properties")
                         display_mode_radio = gr.Radio(["full", "log", "progress"], label="Display Mode", value="full")
+                        rate_unit = gr.Radio(["it/s","s/it"], label="Progress rate unit", value="it/s")
                         bg_color_picker = gr.ColorPicker(label="Background Color", value="#000000")
                         line_numbers_checkbox = gr.Checkbox(label="Show Line Numbers", value=True)
                         autoscroll_checkbox = gr.Checkbox(label="Autoscroll", value=True)
+                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                     with gr.Group():
                         gr.Markdown("### Simulation Controls")
-                        disable_console_checkbox = gr.Checkbox(label="Disable Python Console Output", value=False)
                         start_btn = gr.Button("Run Success Case", variant="primary")
                         error_btn = gr.Button("Run Error Case")
             
-            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox], outputs=feature_logger)
-            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox], outputs=feature_logger)
+            start_btn.click(fn=run_success_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
+            error_btn.click(fn=run_error_case, inputs=[disable_console_checkbox, rate_unit], outputs=feature_logger)
             feature_logger.clear(fn=clear_output, inputs=None, outputs=feature_logger)
             controls = [display_mode_radio, bg_color_picker, line_numbers_checkbox, autoscroll_checkbox]
             for control in controls:
@@ -307,7 +368,7 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
                         label="Result", columns=1, show_label=False, height=500, min_width=768, preview=True, allow_preview=True
                     )
             
-            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
+            run_button.click(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer, run_button])
             prompt.submit(fn=generate, inputs=[prompt], outputs=[result_gallery, livelog_viewer])
             livelog_viewer.clear(fn=clear_output, inputs=None, outputs=[livelog_viewer])
             

src/frontend/shared/LiveLogPanel.svelte

@@ -36,7 +36,7 @@
     // -------------------------------------------------------------------------
 
     /** Holds the current state of the progress bar. */
-    let progress = { visible: true, current: 0, total: 100, desc: "", percentage: 0, rate: 0.0, status: "running" };
+    let progress = { visible: true, current: 0, total: 100, desc: "", percentage: 0, rate: 0.0, status: "running", rate_unit: 'it/s', extra_info:''};
     /** Accumulates all received log lines. */
     let log_lines: { level: string; content: string }[] = [];
     /** A plain text representation of all logs for the utility buttons. */
@@ -66,12 +66,14 @@
         if (value !== null) {
             clearTimeout(debounceTimeout);
             debounceTimeout = setTimeout(async () => {
-                if (value === null) {
-                    // A `null` value is the signal to clear the component's state.
-                    log_lines = [];
-                    progress = { visible: false, current: 0, total: 100, desc: "", percentage: 0, rate: 0.0, status: "running" };
+                if (value === null) {                    
+                    log_lines = [];                    
+                    progress = { 
+                        visible: false, current: 0, total: 100, desc: "", percentage: 0, 
+                        rate: 0.0, status: "running", rate_unit: 'it/s', extra_info: '' 
+                    };
                     all_logs_as_text = "";
-                    initial_desc = "Processing..."; // Reset the initial description
+                    initial_desc = "Processing...";
                 } else if (value) {
                     if (Array.isArray(value)) {
                         // Handles an initial state load if the backend provides a full list.
@@ -84,39 +86,38 @@
                                 progress.visible = true;
                                 progress.current = item.current;
                                 progress.total = item.total || 100;
-                                // Capture the initial fixed description on the first progress update.
                                 if (item.current === 0 && item.desc && initial_desc === "Processing...") {
                                     initial_desc = item.desc;
                                 }
-                                // In 'progress' mode, use the latest log as the description; otherwise, use the fixed one.
                                 progress.desc = display_mode === "progress" && log_lines.length > 0 
                                     ? log_lines[log_lines.length - 1].content 
                                     : initial_desc;
-                                progress.rate = item.rate || 0.0;
+                                progress.rate = item.rate || 0.0;                              
+                                progress.rate_unit = item.rate_unit || 'it/s';
+                                progress.extra_info = item.extra_info || '';
                                 progress.percentage = progress.total > 0 ? ((item.current / progress.total) * 100) : 0;
                                 progress.status = item.status || "running";
                             }
                         }
-                    } else if (typeof value === 'object' && value.type) {
-                        // This is the primary streaming case: handles a single new data object.
+                    } else if (typeof value === 'object' && value.type) {                        
                         if (value.type === "log") {
                             log_lines = [...log_lines, { level: value.level || 'INFO', content: value.content }];
                         } else if (value.type === "progress") {
                             progress.visible = true;
                             progress.current = value.current;
                             progress.total = value.total || 100;
-                            // Capture the initial fixed description on the first progress update.
                             if (value.current === 0 && value.desc && initial_desc === "Processing...") {
                                 initial_desc = value.desc;
                             }
-                            // In 'progress' mode, use the latest log as the description; otherwise, use the fixed one.
                             progress.desc = display_mode === "progress" && log_lines.length > 0 
                                 ? log_lines[log_lines.length - 1].content 
                                 : initial_desc;
-                            progress.rate = value.rate || 0.0;
-                            progress.percentage = progress.total > 0 ? ((value.current / progress.total) * 100) : 0;
+                            progress.rate = value.rate || 0.0;                            
+                            progress.rate_unit = value.rate_unit || 'it/s';
+                            progress.extra_info = value.extra_info || '';
+                            progress.percentage = progress.total > 0 ? ((value.current / value.total) * 100) : 0;
                             progress.status = value.status || "running";
-                            // The backend can send a full log history with a progress update.
+                            
                             log_lines = Array.isArray(value.logs) ? value.logs.map(log => ({
                                 level: log.level || 'INFO',
                                 content: log.content
@@ -125,8 +126,8 @@
                     }
                     all_logs_as_text = log_lines.map(l => l.content).join('\n');
                 }
-                await tick(); // Ensure the DOM is updated before the next potential operation.
-            }, 50); // 50ms debounce window.
+                await tick();
+            }, 50);
         }
     }
 
@@ -166,7 +167,12 @@
         <div class="progress-container">
             <div class="progress-label-top">
                 <span>{progress.desc}</span>
-                <span>{progress.rate.toFixed(1)} it/s</span> 
+                 <span class="rate-info">
+                    {progress.rate.toFixed(2)} {progress.rate_unit}
+                    {#if progress.extra_info}
+                        <span class="extra-info">({progress.extra_info})</span>
+                    {/if}
+                </span> 
             </div>
             <div class="progress-bar-background">
                 <!-- Conditionally apply CSS classes based on the progress status. -->
@@ -282,4 +288,17 @@
     .progress-bar-fill.error {
         background-color: var(--color-error, #ef4444);
     }
+    .rate-info {
+        display: flex;
+        align-items: center;
+        gap: 0.5ch;
+    }
+    .extra-info {
+        color: var(--body-text-color-subdued); 
+        font-size: 0.9em;
+        white-space: nowrap;
+        overflow: hidden;
+        text-overflow: ellipsis;
+        max-width: 200px;
+    }
 </style>

src/pyproject.toml

@@ -8,7 +8,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "gradio_livelog"
-version = "0.0.2"
+version = "0.0.3"
 description = "A Live Log Component for Gradio Interface"
 readme = "README.md"
 license = "apache-2.0"