Update app.py

app.py

@@ -1,13 +1,9 @@
 import gradio as gr
 import spaces
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
-from threading import Thread
-from typing import Iterator
 
 model_name = "rubenroy/Zurich-14B-GCv2-5m"
-MAX_INPUT_TOKEN_LENGTH = 4096
-
 model = AutoModelForCausalLM.from_pretrained(
     model_name,
     torch_dtype=torch.bfloat16,
@@ -16,43 +12,31 @@ model = AutoModelForCausalLM.from_pretrained(
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 
 @spaces.GPU
-def generate(message: str, chat_history: list[tuple[str, str]], temperature=0.7, top_p=0.9, top_k=50, max_new_tokens=512, repetition_penalty=1.1) -> Iterator[str]:
+def generate(message, chat_history, temperature=0.7, top_p=0.9, top_k=50, max_new_tokens=512, repetition_penalty=1.1):
     messages = [
-        {"role": "system", "content": "You are a helpful assistant named Zurich, a 7 billion parameter Large Language Model, fine-tuned and trained by Ruben Roy. You have been trained with the GammaCorpus v2 dataset, a structured and filtered multi-turn conversation dataset created by Ruben Roy."}
+        {"role": "system", "content": "You are a helpul assistant named Zurich, a 14 billion parameter Large Language model, you were fine-tuned and trained by Ruben Roy. You have been trained with the GammaCorpus v2 dataset, a dataset filled with structured and filtered multi-turn conversations, this was also made by Ruben Roy."}, # Attribution to Qwen is not included to prevent hallucinations.
+        {"role": "user", "content": message}
     ]
-    
-    for user, assistant in chat_history:
-        messages.append({"role": "user", "content": user})
-        messages.append({"role": "assistant", "content": assistant})
-    messages.append({"role": "user", "content": message})
-    
-    input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt")
-    
-    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
-        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
-        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
-    
-    input_ids = input_ids.to(model.device)
-    
-    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True, clean_up_tokenization_spaces=True)
-    generate_kwargs = dict(
-        {"input_ids": input_ids},
-        streamer=streamer,
-        max_new_tokens=max_new_tokens,
-        do_sample=True if float(temperature) > 0 else False,
-        top_p=top_p,
-        top_k=top_k,
-        temperature=temperature,
-        repetition_penalty=repetition_penalty
+    text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
     )
-    
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
-    
-    outputs = []
-    for text in streamer:
-        outputs.append(text)
-        yield "".join(outputs)
+    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+    generated_ids = model.generate(
+        **model_inputs,
+        temperature=float(temperature),
+        top_p=float(top_p),
+        top_k=int(top_k),
+        max_new_tokens=int(max_new_tokens),
+        repetition_penalty=float(repetition_penalty),
+        do_sample=True if float(temperature) > 0 else False
+    )
+    generated_ids = [
+        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+    ]
+    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+    return response
 
 TITLE_HTML = """
 <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.0.0/css/all.min.css">
@@ -85,12 +69,85 @@ TITLE_HTML = """
         backdrop-filter: blur(10px);
         transition: all 0.3s ease;
     }
+    .info-link {
+        color: #60a5fa;
+        text-decoration: none;
+        transition: color 0.2s ease;
+    }
+    .info-link:hover {
+        color: #93c5fd;
+        text-decoration: underline;
+    }
+    .info-section {
+        margin-top: 0.5rem;
+        font-size: 0.9rem;
+        color: #94a3b8;
+    }
+    .settings-section {
+        background: rgba(255, 255, 255, 0.05);
+        padding: 1.5rem;
+        border-radius: 1rem;
+        margin: 1.5rem auto;
+        border: 1px solid rgba(255, 255, 255, 0.1);
+        max-width: 800px;
+    }
+    .settings-title {
+        color: #e2e8f0;
+        font-size: 1.25rem;
+        font-weight: 600;
+        margin-bottom: 1rem;
+        display: flex;
+        align-items: center;
+        gap: 0.7rem;
+    }
+    .parameter-info {
+        color: #94a3b8;
+        font-size: 0.8rem;
+        margin-top: 0.25rem;
+    }
 </style>
 
 <div style="background: linear-gradient(135deg, #1e293b 0%, #0f172a 100%); padding: 1.5rem; border-radius: 1.5rem; text-align: center; margin: 1rem auto; max-width: 1200px; box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1);">
     <div style="margin-bottom: 1.5rem;">
-        <h1 style="font-size: 2.5rem; font-weight: 800; margin: 0; background: linear-gradient(135deg, #60a5fa 0%, #93c5fd 100%); -webkit-background-clip: text; -webkit-text-fill-color: transparent;">Zurich</h1>
-        <p style="font-size: 1.25rem; color: #94a3b8; margin: 0;">GammaCorpus v2-5m</p>
+        <div style="display: flex; align-items: center; justify-content: center; gap: 1rem;">
+            <h1 style="font-size: 2.5rem; font-weight: 800; margin: 0; background: linear-gradient(135deg, #60a5fa 0%, #93c5fd 100%); -webkit-background-clip: text; -webkit-text-fill-color: transparent;">Zurich</h1>
+            <div style="width: 2px; height: 2.5rem; background: linear-gradient(180deg, #3b82f6 0%, #60a5fa 100%);"></div>
+            <p style="font-size: 1.25rem; color: #94a3b8; margin: 0;">GammaCorpus v2-5m</p>
+        </div>
+        <div class="info-section">
+            <span>Fine-tuned from <a href="https://huggingface.co/Qwen/Qwen2.5-14B-Instruct" class="info-link">Qwen 2.5 14B Instruct</a> | Model: <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-5m" class="info-link">Zurich-14B-GCv2-5m</a> | Training Dataset: <a href="https://huggingface.co/datasets/rubenroy/GammaCorpus-v2-5m" class="info-link">GammaCorpus v2 5m</a></span>
+        </div>
+    </div>
+
+    <div style="display: flex; gap: 1.5rem; justify-content: center;">
+        <div class="model-section">
+            <h2 style="font-size: 1.25rem; color: #e2e8f0; margin-bottom: 1.4rem; margin-top: 1px; font-weight: 600; display: flex; align-items: center; justify-content: center; gap: 0.7rem;">
+                <i class="fas fa-brain"></i>
+                7B Models
+            </h2>
+            <div style="display: grid; grid-template-columns: repeat(2, 1fr); gap: 0.75rem;">
+                <a href="https://huggingface.co/rubenroy/Zurich-7B-GCv2-5m" class="model-btn">Zurich 7B GCv2 5m</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-7B-GCv2-1m" class="model-btn">Zurich 7B GCv2 1m</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-7B-GCv2-500k" class="model-btn">Zurich 7B GCv2 500k</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-7B-GCv2-100k" class="model-btn">Zurich 7B GCv2 100k</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-7B-GCv2-50k" class="model-btn">Zurich 7B GCv2 50k</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-7B-GCv2-10k" class="model-btn">Zurich 7B GCv2 10k</a>
+            </div>
+        </div>
+        <div class="model-section">
+            <h2 style="font-size: 1.25rem; color: #e2e8f0; margin-bottom: 1.4rem; margin-top: 1px; font-weight: 600; display: flex; align-items: center; justify-content: center; gap: 0.7rem;">
+                <i class="fas fa-rocket"></i>
+                14B Models
+            </h2>
+            <div style="display: grid; grid-template-columns: repeat(2, 1fr); gap: 0.75rem;">
+                <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-5m" class="model-btn">Zurich 14B GCv2 5m</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-1m" class="model-btn">Zurich 14B GCv2 1m</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-500k" class="model-btn">Zurich 14B GCv2 500k</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-100k" class="model-btn">Zurich 14B GCv2 100k</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-50k" class="model-btn">Zurich 14B GCv2 50k</a>
+                <a href="https://huggingface.co/rubenroy/Zurich-14B-GCv2-10k" class="model-btn">Zurich 14B GCv2 10k</a>
+            </div>
+        </div>
     </div>
 </div>
 """
@@ -99,7 +156,7 @@ examples = [
     ["Explain quantum computing in simple terms"],
     ["Write a short story about a time traveler"],
     ["Explain the process of photosynthesis"],
-    ["Tell me an interesting fact about Palm trees"]
+    ["Tell me an intersting fact about Palm trees"]
 ]
 
 with gr.Blocks() as demo:
@@ -108,16 +165,62 @@ with gr.Blocks() as demo:
     with gr.Accordion("Generation Settings", open=False):
         with gr.Row():
             with gr.Column():
-                temperature = gr.Slider(0.0, 2.0, value=0.7, step=0.1, label="Temperature", info="Higher values make the output more random")
-                top_p = gr.Slider(0.0, 1.0, value=0.9, step=0.05, label="Top P", info="Controls nucleus sampling")
-                top_k = gr.Slider(1, 100, value=50, step=1, label="Top K", info="Limits vocabulary choices per step")
+                temperature = gr.Slider(
+                    minimum=0.0,
+                    maximum=2.0,
+                    value=0.7,
+                    step=0.1,
+                    label="Temperature",
+                    info="Higher values make the output more random, lower values make it more deterministic",
+                    interactive=True
+                )
+                top_p = gr.Slider(
+                    minimum=0.0,
+                    maximum=1.0,
+                    value=0.9,
+                    step=0.05,
+                    label="Top P",
+                    info="Controls the cumulative probability threshold for nucleus sampling",
+                    interactive=True
+                )
+                top_k = gr.Slider(
+                    minimum=1,
+                    maximum=100,
+                    value=50,
+                    step=1,
+                    label="Top K",
+                    info="Limits the number of tokens to consider for each generation step",
+                    interactive=True
+                )
             with gr.Column():
-                max_new_tokens = gr.Slider(1, 2048, value=512, step=1, label="Max New Tokens", info="Limits response length")
-                repetition_penalty = gr.Slider(1.0, 2.0, value=1.1, step=0.1, label="Repetition Penalty", info="Discourages repeated phrases")
-
+                max_new_tokens = gr.Slider(
+                    minimum=1,
+                    maximum=2048,
+                    value=512,
+                    step=1,
+                    label="Max New Tokens",
+                    info="Maximum number of tokens to generate in the response",
+                    interactive=True
+                )
+                repetition_penalty = gr.Slider(
+                    minimum=1.0,
+                    maximum=2.0,
+                    value=1.1,
+                    step=0.1,
+                    label="Repetition Penalty",
+                    info="Higher values stop the model from repeating the same info",
+                    interactive=True
+                )
+    
     chatbot = gr.ChatInterface(
         fn=generate,
-        additional_inputs=[temperature, top_p, top_k, max_new_tokens, repetition_penalty],
+        additional_inputs=[
+            temperature,
+            top_p,
+            top_k,
+            max_new_tokens,
+            repetition_penalty
+        ],
         examples=examples
     )