Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -3,7 +3,7 @@
|
|
3 |
import os
|
4 |
import re
|
5 |
import tempfile
|
6 |
-
import gc # garbage collector
|
7 |
from collections.abc import Iterator
|
8 |
from threading import Thread
|
9 |
import json
|
@@ -12,7 +12,7 @@ import cv2
|
|
12 |
import base64
|
13 |
import logging
|
14 |
import time
|
15 |
-
from urllib.parse import quote #
|
16 |
|
17 |
import gradio as gr
|
18 |
import spaces
|
@@ -21,12 +21,12 @@ from loguru import logger
|
|
21 |
from PIL import Image
|
22 |
from transformers import AutoProcessor, Gemma3ForConditionalGeneration, TextIteratorStreamer
|
23 |
|
24 |
-
# CSV/TXT/PDF
|
25 |
import pandas as pd
|
26 |
import PyPDF2
|
27 |
|
28 |
# =============================================================================
|
29 |
-
# (
|
30 |
# =============================================================================
|
31 |
from gradio_client import Client
|
32 |
|
@@ -38,20 +38,20 @@ logging.basicConfig(
|
|
38 |
)
|
39 |
|
40 |
def test_api_connection() -> str:
|
41 |
-
"""API
|
42 |
try:
|
43 |
client = Client(API_URL)
|
44 |
-
return "API
|
45 |
except Exception as e:
|
46 |
-
logging.error(f"API
|
47 |
-
return f"API
|
48 |
|
49 |
def generate_image(prompt: str, width: float, height: float, guidance: float, inference_steps: float, seed: float):
|
50 |
-
"""
|
51 |
if not prompt:
|
52 |
-
return None, "
|
53 |
try:
|
54 |
-
logging.info(f"
|
55 |
|
56 |
client = Client(API_URL)
|
57 |
result = client.predict(
|
@@ -68,32 +68,32 @@ def generate_image(prompt: str, width: float, height: float, guidance: float, in
|
|
68 |
api_name="/generate_image"
|
69 |
)
|
70 |
|
71 |
-
logging.info(f"
|
72 |
|
73 |
-
#
|
74 |
if isinstance(result, (list, tuple)) and len(result) > 0:
|
75 |
-
image_data = result[0] #
|
76 |
-
seed_info = result[1] if len(result) > 1 else "
|
77 |
return image_data, seed_info
|
78 |
else:
|
79 |
-
#
|
80 |
-
return result, "
|
81 |
|
82 |
except Exception as e:
|
83 |
-
logging.error(f"
|
84 |
-
return None, f"
|
85 |
|
86 |
-
# Base64
|
87 |
def fix_base64_padding(data):
|
88 |
-
"""
|
89 |
if isinstance(data, bytes):
|
90 |
data = data.decode('utf-8')
|
91 |
|
92 |
-
#
|
93 |
if "base64," in data:
|
94 |
data = data.split("base64,", 1)[1]
|
95 |
|
96 |
-
#
|
97 |
missing_padding = len(data) % 4
|
98 |
if missing_padding:
|
99 |
data += '=' * (4 - missing_padding)
|
@@ -101,27 +101,27 @@ def fix_base64_padding(data):
|
|
101 |
return data
|
102 |
|
103 |
# =============================================================================
|
104 |
-
#
|
105 |
# =============================================================================
|
106 |
def clear_cuda_cache():
|
107 |
-
"""
|
108 |
if torch.cuda.is_available():
|
109 |
torch.cuda.empty_cache()
|
110 |
gc.collect()
|
111 |
|
112 |
# =============================================================================
|
113 |
-
# SerpHouse
|
114 |
# =============================================================================
|
115 |
SERPHOUSE_API_KEY = os.getenv("SERPHOUSE_API_KEY", "")
|
116 |
|
117 |
def extract_keywords(text: str, top_k: int = 5) -> str:
|
118 |
-
"""
|
119 |
text = re.sub(r"[^a-zA-Z0-9가-힣\s]", "", text)
|
120 |
tokens = text.split()
|
121 |
return " ".join(tokens[:top_k])
|
122 |
|
123 |
def do_web_search(query: str) -> str:
|
124 |
-
"""SerpHouse LIVE API
|
125 |
try:
|
126 |
url = "https://api.serphouse.com/serp/live"
|
127 |
params = {
|
@@ -133,7 +133,7 @@ def do_web_search(query: str) -> str:
|
|
133 |
"num": "20"
|
134 |
}
|
135 |
headers = {"Authorization": f"Bearer {SERPHOUSE_API_KEY}"}
|
136 |
-
logger.info(f"SerpHouse API
|
137 |
response = requests.get(url, headers=headers, params=params, timeout=60)
|
138 |
response.raise_for_status()
|
139 |
data = response.json()
|
@@ -147,38 +147,38 @@ def do_web_search(query: str) -> str:
|
|
147 |
elif "organic" in data:
|
148 |
organic = data["organic"]
|
149 |
if not organic:
|
150 |
-
logger.warning("
|
151 |
-
return "
|
152 |
max_results = min(20, len(organic))
|
153 |
limited_organic = organic[:max_results]
|
154 |
summary_lines = []
|
155 |
for idx, item in enumerate(limited_organic, start=1):
|
156 |
-
title = item.get("title", "
|
157 |
link = item.get("link", "#")
|
158 |
-
snippet = item.get("snippet", "
|
159 |
displayed_link = item.get("displayed_link", link)
|
160 |
summary_lines.append(
|
161 |
-
f"###
|
162 |
f"{snippet}\n\n"
|
163 |
-
f"
|
164 |
f"---\n"
|
165 |
)
|
166 |
instructions = """
|
167 |
-
#
|
168 |
-
|
169 |
-
1.
|
170 |
-
2.
|
171 |
-
3.
|
172 |
-
4.
|
173 |
-
5.
|
174 |
"""
|
175 |
return instructions + "\n".join(summary_lines)
|
176 |
except Exception as e:
|
177 |
-
logger.error(f"
|
178 |
-
return f"
|
179 |
|
180 |
# =============================================================================
|
181 |
-
#
|
182 |
# =============================================================================
|
183 |
MAX_CONTENT_CHARS = 2000
|
184 |
MAX_INPUT_LENGTH = 2096
|
@@ -193,7 +193,7 @@ model = Gemma3ForConditionalGeneration.from_pretrained(
|
|
193 |
MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "5"))
|
194 |
|
195 |
# =============================================================================
|
196 |
-
# CSV, TXT, PDF
|
197 |
# =============================================================================
|
198 |
def analyze_csv_file(path: str) -> str:
|
199 |
try:
|
@@ -202,20 +202,20 @@ def analyze_csv_file(path: str) -> str:
|
|
202 |
df = df.iloc[:50, :10]
|
203 |
df_str = df.to_string()
|
204 |
if len(df_str) > MAX_CONTENT_CHARS:
|
205 |
-
df_str = df_str[:MAX_CONTENT_CHARS] + "\n...(
|
206 |
-
return f"**[CSV
|
207 |
except Exception as e:
|
208 |
-
return f"CSV
|
209 |
|
210 |
def analyze_txt_file(path: str) -> str:
|
211 |
try:
|
212 |
with open(path, "r", encoding="utf-8") as f:
|
213 |
text = f.read()
|
214 |
if len(text) > MAX_CONTENT_CHARS:
|
215 |
-
text = text[:MAX_CONTENT_CHARS] + "\n...(
|
216 |
-
return f"**[TXT
|
217 |
except Exception as e:
|
218 |
-
return f"TXT
|
219 |
|
220 |
def pdf_to_markdown(pdf_path: str) -> str:
|
221 |
text_chunks = []
|
@@ -228,19 +228,19 @@ def pdf_to_markdown(pdf_path: str) -> str:
|
|
228 |
page_text = page_text.strip()
|
229 |
if page_text:
|
230 |
if len(page_text) > MAX_CONTENT_CHARS // max_pages:
|
231 |
-
page_text = page_text[:MAX_CONTENT_CHARS // max_pages] + "...(
|
232 |
-
text_chunks.append(f"##
|
233 |
if len(reader.pages) > max_pages:
|
234 |
-
text_chunks.append(f"\n...(
|
235 |
except Exception as e:
|
236 |
-
return f"PDF
|
237 |
full_text = "\n".join(text_chunks)
|
238 |
if len(full_text) > MAX_CONTENT_CHARS:
|
239 |
-
full_text = full_text[:MAX_CONTENT_CHARS] + "\n...(
|
240 |
-
return f"**[PDF
|
241 |
|
242 |
# =============================================================================
|
243 |
-
#
|
244 |
# =============================================================================
|
245 |
def count_files_in_new_message(paths: list[str]) -> tuple[int, int]:
|
246 |
image_count = 0
|
@@ -274,28 +274,28 @@ def validate_media_constraints(message: dict, history: list[dict]) -> bool:
|
|
274 |
image_count = history_image_count + new_image_count
|
275 |
video_count = history_video_count + new_video_count
|
276 |
if video_count > 1:
|
277 |
-
gr.Warning("
|
278 |
return False
|
279 |
if video_count == 1:
|
280 |
if image_count > 0:
|
281 |
-
gr.Warning("
|
282 |
return False
|
283 |
if "<image>" in message["text"]:
|
284 |
-
gr.Warning("<image>
|
285 |
return False
|
286 |
if video_count == 0 and image_count > MAX_NUM_IMAGES:
|
287 |
-
gr.Warning(f"
|
288 |
return False
|
289 |
if "<image>" in message["text"]:
|
290 |
image_files = [f for f in message["files"] if re.search(r"\.(png|jpg|jpeg|gif|webp)$", f, re.IGNORECASE)]
|
291 |
image_tag_count = message["text"].count("<image>")
|
292 |
if image_tag_count != len(image_files):
|
293 |
-
gr.Warning("
|
294 |
return False
|
295 |
return True
|
296 |
|
297 |
# =============================================================================
|
298 |
-
#
|
299 |
# =============================================================================
|
300 |
def downsample_video(video_path: str) -> list[tuple[Image.Image, float]]:
|
301 |
vidcap = cv2.VideoCapture(video_path)
|
@@ -325,12 +325,12 @@ def process_video(video_path: str) -> tuple[list[dict], list[str]]:
|
|
325 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_file:
|
326 |
pil_image.save(temp_file.name)
|
327 |
temp_files.append(temp_file.name)
|
328 |
-
content.append({"type": "text", "text": f"
|
329 |
content.append({"type": "image", "url": temp_file.name})
|
330 |
return content, temp_files
|
331 |
|
332 |
# =============================================================================
|
333 |
-
#
|
334 |
# =============================================================================
|
335 |
def process_interleaved_images(message: dict) -> list[dict]:
|
336 |
parts = re.split(r"(<image>)", message["text"])
|
@@ -349,7 +349,7 @@ def process_interleaved_images(message: dict) -> list[dict]:
|
|
349 |
return content
|
350 |
|
351 |
# =============================================================================
|
352 |
-
#
|
353 |
# =============================================================================
|
354 |
def is_image_file(file_path: str) -> bool:
|
355 |
return bool(re.search(r"\.(png|jpg|jpeg|gif|webp)$", file_path, re.IGNORECASE))
|
@@ -392,7 +392,7 @@ def process_new_user_message(message: dict) -> tuple[list[dict], list[str]]:
|
|
392 |
return content_list, temp_files
|
393 |
|
394 |
# =============================================================================
|
395 |
-
# history
|
396 |
# =============================================================================
|
397 |
def process_history(history: list[dict]) -> list[dict]:
|
398 |
messages = []
|
@@ -412,24 +412,24 @@ def process_history(history: list[dict]) -> list[dict]:
|
|
412 |
if is_image_file(file_path):
|
413 |
current_user_content.append({"type": "image", "url": file_path})
|
414 |
else:
|
415 |
-
current_user_content.append({"type": "text", "text": f"[
|
416 |
if current_user_content:
|
417 |
messages.append({"role": "user", "content": current_user_content})
|
418 |
return messages
|
419 |
|
420 |
# =============================================================================
|
421 |
-
#
|
422 |
# =============================================================================
|
423 |
def _model_gen_with_oom_catch(**kwargs):
|
424 |
try:
|
425 |
model.generate(**kwargs)
|
426 |
except torch.cuda.OutOfMemoryError:
|
427 |
-
raise RuntimeError("[OutOfMemoryError] GPU
|
428 |
finally:
|
429 |
clear_cuda_cache()
|
430 |
|
431 |
# =============================================================================
|
432 |
-
#
|
433 |
# =============================================================================
|
434 |
@spaces.GPU(duration=120)
|
435 |
def run(
|
@@ -439,43 +439,42 @@ def run(
|
|
439 |
max_new_tokens: int = 512,
|
440 |
use_web_search: bool = False,
|
441 |
web_search_query: str = "",
|
442 |
-
age_group: str = "
|
443 |
mbti_personality: str = "INTP",
|
444 |
sexual_openness: int = 2,
|
445 |
-
image_gen: bool = False # "Image Gen"
|
446 |
) -> Iterator[str]:
|
447 |
if not validate_media_constraints(message, history):
|
448 |
yield ""
|
449 |
return
|
450 |
temp_files = []
|
451 |
try:
|
452 |
-
#
|
453 |
persona = (
|
454 |
f"{system_prompt.strip()}\n\n"
|
455 |
-
f"
|
456 |
-
f"
|
457 |
-
f"MBTI
|
458 |
-
f"
|
459 |
)
|
460 |
-
combined_system_msg = f"[
|
461 |
|
462 |
if use_web_search:
|
463 |
user_text = message["text"]
|
464 |
ws_query = extract_keywords(user_text)
|
465 |
if ws_query.strip():
|
466 |
-
logger.info(f"[
|
467 |
ws_result = do_web_search(ws_query)
|
468 |
-
combined_system_msg += f"[
|
469 |
combined_system_msg += (
|
470 |
-
"[
|
471 |
-
"[
|
472 |
-
"1.
|
473 |
-
"2.
|
474 |
-
"3.
|
475 |
-
"4. 답변 마지막에 \"참고 자료:\" 섹션을 추가하고 사용한 주요 출처 링크를 나열하세요.\n"
|
476 |
)
|
477 |
else:
|
478 |
-
combined_system_msg += "[
|
479 |
messages = []
|
480 |
if combined_system_msg.strip():
|
481 |
messages.append({"role": "system", "content": [{"type": "text", "text": combined_system_msg.strip()}]})
|
@@ -484,7 +483,7 @@ def run(
|
|
484 |
temp_files.extend(user_temp_files)
|
485 |
for item in user_content:
|
486 |
if item["type"] == "text" and len(item["text"]) > MAX_CONTENT_CHARS:
|
487 |
-
item["text"] = item["text"][:MAX_CONTENT_CHARS] + "\n...(
|
488 |
messages.append({"role": "user", "content": user_content})
|
489 |
inputs = processor.apply_chat_template(
|
490 |
messages,
|
@@ -507,16 +506,16 @@ def run(
|
|
507 |
yield output_so_far
|
508 |
|
509 |
except Exception as e:
|
510 |
-
logger.error(f"run
|
511 |
-
yield f"
|
512 |
finally:
|
513 |
for tmp in temp_files:
|
514 |
try:
|
515 |
if os.path.exists(tmp):
|
516 |
os.unlink(tmp)
|
517 |
-
logger.info(f"
|
518 |
except Exception as ee:
|
519 |
-
logger.warning(f"
|
520 |
try:
|
521 |
del inputs, streamer
|
522 |
except Exception:
|
@@ -524,16 +523,16 @@ def run(
|
|
524 |
clear_cuda_cache()
|
525 |
|
526 |
# =============================================================================
|
527 |
-
#
|
528 |
# =============================================================================
|
529 |
def modified_run(message, history, system_prompt, max_new_tokens, use_web_search, web_search_query,
|
530 |
age_group, mbti_personality, sexual_openness, image_gen):
|
531 |
-
#
|
532 |
output_so_far = ""
|
533 |
gallery_update = gr.Gallery(visible=False, value=[])
|
534 |
yield output_so_far, gallery_update
|
535 |
|
536 |
-
#
|
537 |
text_generator = run(message, history, system_prompt, max_new_tokens, use_web_search,
|
538 |
web_search_query, age_group, mbti_personality, sexual_openness, image_gen)
|
539 |
|
@@ -541,15 +540,15 @@ def modified_run(message, history, system_prompt, max_new_tokens, use_web_search
|
|
541 |
output_so_far = text_chunk
|
542 |
yield output_so_far, gallery_update
|
543 |
|
544 |
-
#
|
545 |
if image_gen and message["text"].strip():
|
546 |
try:
|
547 |
width, height = 512, 512
|
548 |
guidance, steps, seed = 7.5, 30, 42
|
549 |
|
550 |
-
logger.info(f"
|
551 |
|
552 |
-
# API
|
553 |
image_result, seed_info = generate_image(
|
554 |
prompt=message["text"].strip(),
|
555 |
width=width,
|
@@ -560,133 +559,120 @@ def modified_run(message, history, system_prompt, max_new_tokens, use_web_search
|
|
560 |
)
|
561 |
|
562 |
if image_result:
|
563 |
-
#
|
564 |
if isinstance(image_result, str) and (
|
565 |
image_result.startswith('data:') or
|
566 |
-
len(image_result) > 100 and '/' not in image_result
|
567 |
):
|
568 |
-
# base64 이미지 문자열을 파일로 변환
|
569 |
try:
|
570 |
-
# data:image
|
571 |
if image_result.startswith('data:'):
|
572 |
content_type, b64data = image_result.split(';base64,')
|
573 |
else:
|
574 |
b64data = image_result
|
575 |
-
content_type = "image/webp" #
|
576 |
|
577 |
-
# base64
|
578 |
image_bytes = base64.b64decode(b64data)
|
579 |
|
580 |
-
#
|
581 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
582 |
temp_file.write(image_bytes)
|
583 |
temp_path = temp_file.name
|
584 |
|
585 |
-
#
|
586 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
587 |
-
yield output_so_far + "\n\n
|
588 |
|
589 |
except Exception as e:
|
590 |
-
logger.error(f"
|
591 |
-
yield output_so_far + f"\n\n(
|
592 |
|
593 |
-
#
|
594 |
elif isinstance(image_result, str) and os.path.exists(image_result):
|
595 |
-
# 로컬 파일 경로를 그대로 사용
|
596 |
gallery_update = gr.Gallery(visible=True, value=[image_result])
|
597 |
-
yield output_so_far + "\n\n
|
598 |
|
599 |
-
# /tmp
|
600 |
elif isinstance(image_result, str) and '/tmp/' in image_result:
|
601 |
-
# API에서 반환된 파일 경로에서 이미지 정보 추출
|
602 |
try:
|
603 |
-
# API 응답을 base64 인코딩된 문자열로 처리
|
604 |
client = Client(API_URL)
|
605 |
result = client.predict(
|
606 |
prompt=message["text"].strip(),
|
607 |
-
api_name="/generate_base64_image" #
|
608 |
)
|
609 |
|
610 |
if isinstance(result, str) and (result.startswith('data:') or len(result) > 100):
|
611 |
-
# base64 이미지 처리
|
612 |
if result.startswith('data:'):
|
613 |
content_type, b64data = result.split(';base64,')
|
614 |
else:
|
615 |
b64data = result
|
616 |
|
617 |
-
# base64 디코딩
|
618 |
image_bytes = base64.b64decode(b64data)
|
619 |
|
620 |
-
# 임시 파일로 저장
|
621 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
622 |
temp_file.write(image_bytes)
|
623 |
temp_path = temp_file.name
|
624 |
|
625 |
-
# 갤러리 표시 및 이미지 추가
|
626 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
627 |
-
yield output_so_far + "\n\n
|
628 |
else:
|
629 |
-
yield output_so_far + "\n\n(
|
630 |
|
631 |
except Exception as e:
|
632 |
-
logger.error(f"
|
633 |
-
yield output_so_far + f"\n\n(
|
634 |
|
635 |
-
# URL
|
636 |
elif isinstance(image_result, str) and (
|
637 |
image_result.startswith('http://') or
|
638 |
image_result.startswith('https://')
|
639 |
):
|
640 |
try:
|
641 |
-
# URL에서 이미지 다운로드
|
642 |
response = requests.get(image_result, timeout=10)
|
643 |
response.raise_for_status()
|
644 |
|
645 |
-
# 임시 파일로 저장
|
646 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
647 |
temp_file.write(response.content)
|
648 |
temp_path = temp_file.name
|
649 |
|
650 |
-
# 갤러리 표시 및 이미지 추가
|
651 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
652 |
-
yield output_so_far + "\n\n
|
653 |
|
654 |
except Exception as e:
|
655 |
-
logger.error(f"URL
|
656 |
-
yield output_so_far + f"\n\n(
|
657 |
|
658 |
-
#
|
659 |
elif hasattr(image_result, 'save'):
|
660 |
try:
|
661 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
662 |
image_result.save(temp_file.name)
|
663 |
temp_path = temp_file.name
|
664 |
|
665 |
-
# 갤러리 표시 및 이미지 추가
|
666 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
667 |
-
yield output_so_far + "\n\n
|
668 |
|
669 |
except Exception as e:
|
670 |
-
logger.error(f"
|
671 |
-
yield output_so_far + f"\n\n(
|
672 |
|
673 |
else:
|
674 |
-
|
675 |
-
yield output_so_far + f"\n\n(지원되지 않는 이미지 형식: {type(image_result)})", gallery_update
|
676 |
else:
|
677 |
-
yield output_so_far + f"\n\n(
|
678 |
|
679 |
except Exception as e:
|
680 |
-
logger.error(f"
|
681 |
-
yield output_so_far + f"\n\n(
|
682 |
|
683 |
# =============================================================================
|
684 |
-
#
|
685 |
# =============================================================================
|
686 |
examples = [
|
687 |
[
|
688 |
{
|
689 |
-
"text": "
|
690 |
"files": [
|
691 |
"assets/additional-examples/before.pdf",
|
692 |
"assets/additional-examples/after.pdf",
|
@@ -695,25 +681,25 @@ examples = [
|
|
695 |
],
|
696 |
[
|
697 |
{
|
698 |
-
"text": "
|
699 |
"files": ["assets/additional-examples/sample-csv.csv"],
|
700 |
}
|
701 |
],
|
702 |
[
|
703 |
{
|
704 |
-
"text": "
|
705 |
"files": ["assets/additional-examples/tmp.mp4"],
|
706 |
}
|
707 |
],
|
708 |
[
|
709 |
{
|
710 |
-
"text": "
|
711 |
"files": ["assets/additional-examples/maz.jpg"],
|
712 |
}
|
713 |
],
|
714 |
[
|
715 |
{
|
716 |
-
"text": "
|
717 |
"files": [
|
718 |
"assets/additional-examples/pill1.png",
|
719 |
"assets/additional-examples/pill2.png"
|
@@ -722,19 +708,19 @@ examples = [
|
|
722 |
],
|
723 |
[
|
724 |
{
|
725 |
-
"text": "
|
726 |
"files": ["assets/additional-examples/4.png"],
|
727 |
}
|
728 |
],
|
729 |
[
|
730 |
{
|
731 |
-
"text": "
|
732 |
"files": ["assets/additional-examples/2.png"],
|
733 |
}
|
734 |
],
|
735 |
[
|
736 |
{
|
737 |
-
"text": "
|
738 |
"files": [
|
739 |
"assets/sample-images/09-1.png",
|
740 |
"assets/sample-images/09-2.png",
|
@@ -746,36 +732,36 @@ examples = [
|
|
746 |
],
|
747 |
[
|
748 |
{
|
749 |
-
"text": "
|
750 |
"files": ["assets/additional-examples/barchart.png"],
|
751 |
}
|
752 |
],
|
753 |
[
|
754 |
{
|
755 |
-
"text": "
|
756 |
"files": ["assets/additional-examples/3.png"],
|
757 |
}
|
758 |
],
|
759 |
|
760 |
[
|
761 |
{
|
762 |
-
"text": "
|
763 |
"files": ["assets/sample-images/03.png"],
|
764 |
}
|
765 |
],
|
766 |
[
|
767 |
{
|
768 |
-
"text": "
|
769 |
}
|
770 |
],
|
771 |
|
772 |
]
|
773 |
|
774 |
# =============================================================================
|
775 |
-
# Gradio UI (Blocks)
|
776 |
# =============================================================================
|
777 |
|
778 |
-
# 1. Gradio Blocks UI
|
779 |
css = """
|
780 |
.gradio-container {
|
781 |
background: rgba(255, 255, 255, 0.7);
|
@@ -786,19 +772,19 @@ css = """
|
|
786 |
}
|
787 |
"""
|
788 |
title_html = """
|
789 |
-
<h1 align="center" style="margin-bottom: 0.2em; font-size: 1.6em;"> 💘 HeartSync Korea💘 </h1>
|
790 |
<p align="center" style="font-size:1.1em; color:#555;">
|
791 |
-
ChatGPT-4o
|
792 |
-
✅FLUX
|
793 |
</p>
|
794 |
"""
|
795 |
|
796 |
with gr.Blocks(css=css, title="AgenticAI-Kv1") as demo:
|
797 |
gr.Markdown(title_html)
|
798 |
|
799 |
-
#
|
800 |
generated_images = gr.Gallery(
|
801 |
-
label="
|
802 |
show_label=True,
|
803 |
visible=False,
|
804 |
elem_id="generated_images",
|
@@ -808,67 +794,70 @@ with gr.Blocks(css=css, title="AgenticAI-Kv1") as demo:
|
|
808 |
)
|
809 |
|
810 |
with gr.Row():
|
811 |
-
web_search_checkbox = gr.Checkbox(label="
|
812 |
-
image_gen_checkbox = gr.Checkbox(label="
|
813 |
|
814 |
base_system_prompt_box = gr.Textbox(
|
815 |
lines=3,
|
816 |
-
value="
|
817 |
-
|
|
|
|
|
|
|
818 |
visible=False
|
819 |
)
|
820 |
with gr.Row():
|
821 |
age_group_dropdown = gr.Dropdown(
|
822 |
-
label="
|
823 |
-
choices=["
|
824 |
-
value="
|
825 |
interactive=True
|
826 |
)
|
827 |
-
# MBTI
|
828 |
mbti_choices = [
|
829 |
-
"INTJ (
|
830 |
-
"INTP (
|
831 |
-
"ENTJ (
|
832 |
-
"ENTP (
|
833 |
-
"INFJ (
|
834 |
-
"INFP (
|
835 |
-
"ENFJ (
|
836 |
-
"ENFP (
|
837 |
-
"ISTJ (
|
838 |
-
"ISFJ (
|
839 |
-
"ESTJ (
|
840 |
-
"ESFJ (
|
841 |
-
"ISTP (
|
842 |
-
"ISFP (
|
843 |
-
"ESTP (
|
844 |
-
"ESFP (
|
845 |
]
|
846 |
mbti_dropdown = gr.Dropdown(
|
847 |
-
label="AI
|
848 |
choices=mbti_choices,
|
849 |
-
value="INTP (
|
850 |
interactive=True
|
851 |
)
|
852 |
sexual_openness_slider = gr.Slider(
|
853 |
minimum=1, maximum=5, step=1, value=2,
|
854 |
-
label="
|
855 |
interactive=True
|
856 |
)
|
857 |
max_tokens_slider = gr.Slider(
|
858 |
-
label="
|
859 |
minimum=100, maximum=8000, step=50, value=1000,
|
860 |
visible=False
|
861 |
)
|
862 |
web_search_text = gr.Textbox(
|
863 |
lines=1,
|
864 |
-
label="
|
865 |
-
placeholder="
|
866 |
visible=False
|
867 |
)
|
868 |
|
869 |
-
#
|
870 |
chat = gr.ChatInterface(
|
871 |
-
fn=modified_run, #
|
872 |
type="messages",
|
873 |
chatbot=gr.Chatbot(type="messages", scale=1, allow_tags=["image"]),
|
874 |
textbox=gr.MultimodalTextbox(
|
@@ -888,7 +877,7 @@ with gr.Blocks(css=css, title="AgenticAI-Kv1") as demo:
|
|
888 |
image_gen_checkbox,
|
889 |
],
|
890 |
additional_outputs=[
|
891 |
-
generated_images, #
|
892 |
],
|
893 |
stop_btn=False,
|
894 |
# title='<a href="https://discord.gg/openfreeai" target="_blank">https://discord.gg/openfreeai</a>',
|
@@ -902,7 +891,7 @@ with gr.Blocks(css=css, title="AgenticAI-Kv1") as demo:
|
|
902 |
|
903 |
with gr.Row(elem_id="examples_row"):
|
904 |
with gr.Column(scale=12, elem_id="examples_container"):
|
905 |
-
gr.Markdown("###
|
906 |
|
907 |
if __name__ == "__main__":
|
908 |
demo.launch(share=True)
|
|
|
3 |
import os
|
4 |
import re
|
5 |
import tempfile
|
6 |
+
import gc # Added garbage collector
|
7 |
from collections.abc import Iterator
|
8 |
from threading import Thread
|
9 |
import json
|
|
|
12 |
import base64
|
13 |
import logging
|
14 |
import time
|
15 |
+
from urllib.parse import quote # Added for URL encoding
|
16 |
|
17 |
import gradio as gr
|
18 |
import spaces
|
|
|
21 |
from PIL import Image
|
22 |
from transformers import AutoProcessor, Gemma3ForConditionalGeneration, TextIteratorStreamer
|
23 |
|
24 |
+
# CSV/TXT/PDF analysis
|
25 |
import pandas as pd
|
26 |
import PyPDF2
|
27 |
|
28 |
# =============================================================================
|
29 |
+
# (New) Image API related functions
|
30 |
# =============================================================================
|
31 |
from gradio_client import Client
|
32 |
|
|
|
38 |
)
|
39 |
|
40 |
def test_api_connection() -> str:
|
41 |
+
"""Test API server connection"""
|
42 |
try:
|
43 |
client = Client(API_URL)
|
44 |
+
return "API connection successful: Operating normally"
|
45 |
except Exception as e:
|
46 |
+
logging.error(f"API connection test failed: {e}")
|
47 |
+
return f"API connection failed: {e}"
|
48 |
|
49 |
def generate_image(prompt: str, width: float, height: float, guidance: float, inference_steps: float, seed: float):
|
50 |
+
"""Image generation function (flexible return types)"""
|
51 |
if not prompt:
|
52 |
+
return None, "Error: A prompt is required."
|
53 |
try:
|
54 |
+
logging.info(f"Calling image generation API with prompt: {prompt}")
|
55 |
|
56 |
client = Client(API_URL)
|
57 |
result = client.predict(
|
|
|
68 |
api_name="/generate_image"
|
69 |
)
|
70 |
|
71 |
+
logging.info(f"Image generation result: {type(result)}, length: {len(result) if isinstance(result, (list, tuple)) else 'unknown'}")
|
72 |
|
73 |
+
# Handle cases where the result is a tuple or list
|
74 |
if isinstance(result, (list, tuple)) and len(result) > 0:
|
75 |
+
image_data = result[0] # The first element is the image data
|
76 |
+
seed_info = result[1] if len(result) > 1 else "Unknown seed"
|
77 |
return image_data, seed_info
|
78 |
else:
|
79 |
+
# When a single value is returned
|
80 |
+
return result, "Unknown seed"
|
81 |
|
82 |
except Exception as e:
|
83 |
+
logging.error(f"Image generation failed: {str(e)}")
|
84 |
+
return None, f"Error: {str(e)}"
|
85 |
|
86 |
+
# Base64 padding fix function
|
87 |
def fix_base64_padding(data):
|
88 |
+
"""Fix the padding of a Base64 string."""
|
89 |
if isinstance(data, bytes):
|
90 |
data = data.decode('utf-8')
|
91 |
|
92 |
+
# Remove the prefix if present
|
93 |
if "base64," in data:
|
94 |
data = data.split("base64,", 1)[1]
|
95 |
|
96 |
+
# Add padding characters (to make the length a multiple of 4)
|
97 |
missing_padding = len(data) % 4
|
98 |
if missing_padding:
|
99 |
data += '=' * (4 - missing_padding)
|
|
|
101 |
return data
|
102 |
|
103 |
# =============================================================================
|
104 |
+
# Memory cleanup function
|
105 |
# =============================================================================
|
106 |
def clear_cuda_cache():
|
107 |
+
"""Explicitly clear the CUDA cache."""
|
108 |
if torch.cuda.is_available():
|
109 |
torch.cuda.empty_cache()
|
110 |
gc.collect()
|
111 |
|
112 |
# =============================================================================
|
113 |
+
# SerpHouse related functions
|
114 |
# =============================================================================
|
115 |
SERPHOUSE_API_KEY = os.getenv("SERPHOUSE_API_KEY", "")
|
116 |
|
117 |
def extract_keywords(text: str, top_k: int = 5) -> str:
|
118 |
+
"""Simple keyword extraction: only keep English, Korean, numbers, and spaces."""
|
119 |
text = re.sub(r"[^a-zA-Z0-9가-힣\s]", "", text)
|
120 |
tokens = text.split()
|
121 |
return " ".join(tokens[:top_k])
|
122 |
|
123 |
def do_web_search(query: str) -> str:
|
124 |
+
"""Call the SerpHouse LIVE API to return Markdown formatted search results"""
|
125 |
try:
|
126 |
url = "https://api.serphouse.com/serp/live"
|
127 |
params = {
|
|
|
133 |
"num": "20"
|
134 |
}
|
135 |
headers = {"Authorization": f"Bearer {SERPHOUSE_API_KEY}"}
|
136 |
+
logger.info(f"Calling SerpHouse API with query: {query}")
|
137 |
response = requests.get(url, headers=headers, params=params, timeout=60)
|
138 |
response.raise_for_status()
|
139 |
data = response.json()
|
|
|
147 |
elif "organic" in data:
|
148 |
organic = data["organic"]
|
149 |
if not organic:
|
150 |
+
logger.warning("Organic results not found in response.")
|
151 |
+
return "No web search results available or the API response structure is unexpected."
|
152 |
max_results = min(20, len(organic))
|
153 |
limited_organic = organic[:max_results]
|
154 |
summary_lines = []
|
155 |
for idx, item in enumerate(limited_organic, start=1):
|
156 |
+
title = item.get("title", "No Title")
|
157 |
link = item.get("link", "#")
|
158 |
+
snippet = item.get("snippet", "No Description")
|
159 |
displayed_link = item.get("displayed_link", link)
|
160 |
summary_lines.append(
|
161 |
+
f"### Result {idx}: {title}\n\n"
|
162 |
f"{snippet}\n\n"
|
163 |
+
f"**Source**: [{displayed_link}]({link})\n\n"
|
164 |
f"---\n"
|
165 |
)
|
166 |
instructions = """
|
167 |
+
# Web Search Results
|
168 |
+
Below are the search results. Use this information to answer the query:
|
169 |
+
1. Refer to each result's title, description, and source link.
|
170 |
+
2. In your answer, explicitly cite the source of any used information (e.g., "[Source Title](link)").
|
171 |
+
3. Include the actual source links in your response.
|
172 |
+
4. Synthesize information from multiple sources.
|
173 |
+
5. At the end include a "References:" section listing the main source links.
|
174 |
"""
|
175 |
return instructions + "\n".join(summary_lines)
|
176 |
except Exception as e:
|
177 |
+
logger.error(f"Web search failed: {e}")
|
178 |
+
return f"Web search failed: {str(e)}"
|
179 |
|
180 |
# =============================================================================
|
181 |
+
# Model and processor loading
|
182 |
# =============================================================================
|
183 |
MAX_CONTENT_CHARS = 2000
|
184 |
MAX_INPUT_LENGTH = 2096
|
|
|
193 |
MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "5"))
|
194 |
|
195 |
# =============================================================================
|
196 |
+
# CSV, TXT, PDF analysis functions
|
197 |
# =============================================================================
|
198 |
def analyze_csv_file(path: str) -> str:
|
199 |
try:
|
|
|
202 |
df = df.iloc[:50, :10]
|
203 |
df_str = df.to_string()
|
204 |
if len(df_str) > MAX_CONTENT_CHARS:
|
205 |
+
df_str = df_str[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
206 |
+
return f"**[CSV File: {os.path.basename(path)}]**\n\n{df_str}"
|
207 |
except Exception as e:
|
208 |
+
return f"CSV file read failed ({os.path.basename(path)}): {str(e)}"
|
209 |
|
210 |
def analyze_txt_file(path: str) -> str:
|
211 |
try:
|
212 |
with open(path, "r", encoding="utf-8") as f:
|
213 |
text = f.read()
|
214 |
if len(text) > MAX_CONTENT_CHARS:
|
215 |
+
text = text[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
216 |
+
return f"**[TXT File: {os.path.basename(path)}]**\n\n{text}"
|
217 |
except Exception as e:
|
218 |
+
return f"TXT file read failed ({os.path.basename(path)}): {str(e)}"
|
219 |
|
220 |
def pdf_to_markdown(pdf_path: str) -> str:
|
221 |
text_chunks = []
|
|
|
228 |
page_text = page_text.strip()
|
229 |
if page_text:
|
230 |
if len(page_text) > MAX_CONTENT_CHARS // max_pages:
|
231 |
+
page_text = page_text[:MAX_CONTENT_CHARS // max_pages] + "...(truncated)"
|
232 |
+
text_chunks.append(f"## Page {page_num+1}\n\n{page_text}\n")
|
233 |
if len(reader.pages) > max_pages:
|
234 |
+
text_chunks.append(f"\n...(Displaying only {max_pages} out of {len(reader.pages)} pages)...")
|
235 |
except Exception as e:
|
236 |
+
return f"PDF file read failed ({os.path.basename(pdf_path)}): {str(e)}"
|
237 |
full_text = "\n".join(text_chunks)
|
238 |
if len(full_text) > MAX_CONTENT_CHARS:
|
239 |
+
full_text = full_text[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
240 |
+
return f"**[PDF File: {os.path.basename(pdf_path)}]**\n\n{full_text}"
|
241 |
|
242 |
# =============================================================================
|
243 |
+
# Check media file limits
|
244 |
# =============================================================================
|
245 |
def count_files_in_new_message(paths: list[str]) -> tuple[int, int]:
|
246 |
image_count = 0
|
|
|
274 |
image_count = history_image_count + new_image_count
|
275 |
video_count = history_video_count + new_video_count
|
276 |
if video_count > 1:
|
277 |
+
gr.Warning("Only one video file is supported.")
|
278 |
return False
|
279 |
if video_count == 1:
|
280 |
if image_count > 0:
|
281 |
+
gr.Warning("Mixing images and a video is not allowed.")
|
282 |
return False
|
283 |
if "<image>" in message["text"]:
|
284 |
+
gr.Warning("The <image> tag cannot be used together with a video file.")
|
285 |
return False
|
286 |
if video_count == 0 and image_count > MAX_NUM_IMAGES:
|
287 |
+
gr.Warning(f"You can upload a maximum of {MAX_NUM_IMAGES} images.")
|
288 |
return False
|
289 |
if "<image>" in message["text"]:
|
290 |
image_files = [f for f in message["files"] if re.search(r"\.(png|jpg|jpeg|gif|webp)$", f, re.IGNORECASE)]
|
291 |
image_tag_count = message["text"].count("<image>")
|
292 |
if image_tag_count != len(image_files):
|
293 |
+
gr.Warning("The number of <image> tags does not match the number of image files provided.")
|
294 |
return False
|
295 |
return True
|
296 |
|
297 |
# =============================================================================
|
298 |
+
# Video processing functions
|
299 |
# =============================================================================
|
300 |
def downsample_video(video_path: str) -> list[tuple[Image.Image, float]]:
|
301 |
vidcap = cv2.VideoCapture(video_path)
|
|
|
325 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_file:
|
326 |
pil_image.save(temp_file.name)
|
327 |
temp_files.append(temp_file.name)
|
328 |
+
content.append({"type": "text", "text": f"Frame {timestamp}:"})
|
329 |
content.append({"type": "image", "url": temp_file.name})
|
330 |
return content, temp_files
|
331 |
|
332 |
# =============================================================================
|
333 |
+
# Interleaved <image> processing function
|
334 |
# =============================================================================
|
335 |
def process_interleaved_images(message: dict) -> list[dict]:
|
336 |
parts = re.split(r"(<image>)", message["text"])
|
|
|
349 |
return content
|
350 |
|
351 |
# =============================================================================
|
352 |
+
# File processing -> content creation
|
353 |
# =============================================================================
|
354 |
def is_image_file(file_path: str) -> bool:
|
355 |
return bool(re.search(r"\.(png|jpg|jpeg|gif|webp)$", file_path, re.IGNORECASE))
|
|
|
392 |
return content_list, temp_files
|
393 |
|
394 |
# =============================================================================
|
395 |
+
# Convert history to LLM messages
|
396 |
# =============================================================================
|
397 |
def process_history(history: list[dict]) -> list[dict]:
|
398 |
messages = []
|
|
|
412 |
if is_image_file(file_path):
|
413 |
current_user_content.append({"type": "image", "url": file_path})
|
414 |
else:
|
415 |
+
current_user_content.append({"type": "text", "text": f"[File: {os.path.basename(file_path)}]"})
|
416 |
if current_user_content:
|
417 |
messages.append({"role": "user", "content": current_user_content})
|
418 |
return messages
|
419 |
|
420 |
# =============================================================================
|
421 |
+
# Model generation function (with OOM catching)
|
422 |
# =============================================================================
|
423 |
def _model_gen_with_oom_catch(**kwargs):
|
424 |
try:
|
425 |
model.generate(**kwargs)
|
426 |
except torch.cuda.OutOfMemoryError:
|
427 |
+
raise RuntimeError("[OutOfMemoryError] Insufficient GPU memory.")
|
428 |
finally:
|
429 |
clear_cuda_cache()
|
430 |
|
431 |
# =============================================================================
|
432 |
+
# Main inference function
|
433 |
# =============================================================================
|
434 |
@spaces.GPU(duration=120)
|
435 |
def run(
|
|
|
439 |
max_new_tokens: int = 512,
|
440 |
use_web_search: bool = False,
|
441 |
web_search_query: str = "",
|
442 |
+
age_group: str = "20s",
|
443 |
mbti_personality: str = "INTP",
|
444 |
sexual_openness: int = 2,
|
445 |
+
image_gen: bool = False # "Image Gen" checkbox status
|
446 |
) -> Iterator[str]:
|
447 |
if not validate_media_constraints(message, history):
|
448 |
yield ""
|
449 |
return
|
450 |
temp_files = []
|
451 |
try:
|
452 |
+
# Append persona information to the system prompt
|
453 |
persona = (
|
454 |
f"{system_prompt.strip()}\n\n"
|
455 |
+
f"Gender: Female\n"
|
456 |
+
f"Age Group: {age_group}\n"
|
457 |
+
f"MBTI Persona: {mbti_personality}\n"
|
458 |
+
f"Sexual Openness (1-5): {sexual_openness}\n"
|
459 |
)
|
460 |
+
combined_system_msg = f"[System Prompt]\n{persona.strip()}\n\n"
|
461 |
|
462 |
if use_web_search:
|
463 |
user_text = message["text"]
|
464 |
ws_query = extract_keywords(user_text)
|
465 |
if ws_query.strip():
|
466 |
+
logger.info(f"[Auto web search keywords] {ws_query!r}")
|
467 |
ws_result = do_web_search(ws_query)
|
468 |
+
combined_system_msg += f"[Search Results (Top 20 Items)]\n{ws_result}\n\n"
|
469 |
combined_system_msg += (
|
470 |
+
"[Note: In your answer, cite the above search result links as sources]\n"
|
471 |
+
"[Important Instructions]\n"
|
472 |
+
"1. Include a citation in the format \"[Source Title](link)\" for any information from the search results.\n"
|
473 |
+
"2. Synthesize information from multiple sources when answering.\n"
|
474 |
+
"3. At the end, add a \"References:\" section listing the main source links.\n"
|
|
|
475 |
)
|
476 |
else:
|
477 |
+
combined_system_msg += "[No valid keywords found; skipping web search]\n\n"
|
478 |
messages = []
|
479 |
if combined_system_msg.strip():
|
480 |
messages.append({"role": "system", "content": [{"type": "text", "text": combined_system_msg.strip()}]})
|
|
|
483 |
temp_files.extend(user_temp_files)
|
484 |
for item in user_content:
|
485 |
if item["type"] == "text" and len(item["text"]) > MAX_CONTENT_CHARS:
|
486 |
+
item["text"] = item["text"][:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
487 |
messages.append({"role": "user", "content": user_content})
|
488 |
inputs = processor.apply_chat_template(
|
489 |
messages,
|
|
|
506 |
yield output_so_far
|
507 |
|
508 |
except Exception as e:
|
509 |
+
logger.error(f"Error in run function: {str(e)}")
|
510 |
+
yield f"Sorry, an error occurred: {str(e)}"
|
511 |
finally:
|
512 |
for tmp in temp_files:
|
513 |
try:
|
514 |
if os.path.exists(tmp):
|
515 |
os.unlink(tmp)
|
516 |
+
logger.info(f"Temporary file deleted: {tmp}")
|
517 |
except Exception as ee:
|
518 |
+
logger.warning(f"Failed to delete temporary file {tmp}: {ee}")
|
519 |
try:
|
520 |
del inputs, streamer
|
521 |
except Exception:
|
|
|
523 |
clear_cuda_cache()
|
524 |
|
525 |
# =============================================================================
|
526 |
+
# Modified model run function - handles image generation and gallery update
|
527 |
# =============================================================================
|
528 |
def modified_run(message, history, system_prompt, max_new_tokens, use_web_search, web_search_query,
|
529 |
age_group, mbti_personality, sexual_openness, image_gen):
|
530 |
+
# Initialize and hide the gallery component
|
531 |
output_so_far = ""
|
532 |
gallery_update = gr.Gallery(visible=False, value=[])
|
533 |
yield output_so_far, gallery_update
|
534 |
|
535 |
+
# Execute the original run function
|
536 |
text_generator = run(message, history, system_prompt, max_new_tokens, use_web_search,
|
537 |
web_search_query, age_group, mbti_personality, sexual_openness, image_gen)
|
538 |
|
|
|
540 |
output_so_far = text_chunk
|
541 |
yield output_so_far, gallery_update
|
542 |
|
543 |
+
# If image generation is enabled and there is text input, update the gallery
|
544 |
if image_gen and message["text"].strip():
|
545 |
try:
|
546 |
width, height = 512, 512
|
547 |
guidance, steps, seed = 7.5, 30, 42
|
548 |
|
549 |
+
logger.info(f"Calling image generation for gallery with prompt: {message['text']}")
|
550 |
|
551 |
+
# Call the API to generate an image
|
552 |
image_result, seed_info = generate_image(
|
553 |
prompt=message["text"].strip(),
|
554 |
width=width,
|
|
|
559 |
)
|
560 |
|
561 |
if image_result:
|
562 |
+
# Process image data directly if it is a base64 string
|
563 |
if isinstance(image_result, str) and (
|
564 |
image_result.startswith('data:') or
|
565 |
+
(len(image_result) > 100 and '/' not in image_result)
|
566 |
):
|
|
|
567 |
try:
|
568 |
+
# Remove the data:image prefix if present
|
569 |
if image_result.startswith('data:'):
|
570 |
content_type, b64data = image_result.split(';base64,')
|
571 |
else:
|
572 |
b64data = image_result
|
573 |
+
content_type = "image/webp" # Assume default
|
574 |
|
575 |
+
# Decode base64
|
576 |
image_bytes = base64.b64decode(b64data)
|
577 |
|
578 |
+
# Save to a temporary file
|
579 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
580 |
temp_file.write(image_bytes)
|
581 |
temp_path = temp_file.name
|
582 |
|
583 |
+
# Update gallery to show the image
|
584 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
585 |
+
yield output_so_far + "\n\n*Image generated and displayed in the gallery below.*", gallery_update
|
586 |
|
587 |
except Exception as e:
|
588 |
+
logger.error(f"Error processing Base64 image: {e}")
|
589 |
+
yield output_so_far + f"\n\n(Error processing image: {e})", gallery_update
|
590 |
|
591 |
+
# If the result is a file path
|
592 |
elif isinstance(image_result, str) and os.path.exists(image_result):
|
|
|
593 |
gallery_update = gr.Gallery(visible=True, value=[image_result])
|
594 |
+
yield output_so_far + "\n\n*Image generated and displayed in the gallery below.*", gallery_update
|
595 |
|
596 |
+
# If the path is from /tmp (only on the API server)
|
597 |
elif isinstance(image_result, str) and '/tmp/' in image_result:
|
|
|
598 |
try:
|
|
|
599 |
client = Client(API_URL)
|
600 |
result = client.predict(
|
601 |
prompt=message["text"].strip(),
|
602 |
+
api_name="/generate_base64_image" # API that returns base64
|
603 |
)
|
604 |
|
605 |
if isinstance(result, str) and (result.startswith('data:') or len(result) > 100):
|
|
|
606 |
if result.startswith('data:'):
|
607 |
content_type, b64data = result.split(';base64,')
|
608 |
else:
|
609 |
b64data = result
|
610 |
|
|
|
611 |
image_bytes = base64.b64decode(b64data)
|
612 |
|
|
|
613 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
614 |
temp_file.write(image_bytes)
|
615 |
temp_path = temp_file.name
|
616 |
|
|
|
617 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
618 |
+
yield output_so_far + "\n\n*Image generated and displayed in the gallery below.*", gallery_update
|
619 |
else:
|
620 |
+
yield output_so_far + "\n\n(Image generation failed: Invalid format)", gallery_update
|
621 |
|
622 |
except Exception as e:
|
623 |
+
logger.error(f"Error calling alternative API: {e}")
|
624 |
+
yield output_so_far + f"\n\n(Image generation failed: {e})", gallery_update
|
625 |
|
626 |
+
# If the result is a URL
|
627 |
elif isinstance(image_result, str) and (
|
628 |
image_result.startswith('http://') or
|
629 |
image_result.startswith('https://')
|
630 |
):
|
631 |
try:
|
|
|
632 |
response = requests.get(image_result, timeout=10)
|
633 |
response.raise_for_status()
|
634 |
|
|
|
635 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
636 |
temp_file.write(response.content)
|
637 |
temp_path = temp_file.name
|
638 |
|
|
|
639 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
640 |
+
yield output_so_far + "\n\n*Image generated and displayed in the gallery below.*", gallery_update
|
641 |
|
642 |
except Exception as e:
|
643 |
+
logger.error(f"URL image download error: {e}")
|
644 |
+
yield output_so_far + f"\n\n(Error downloading image: {e})", gallery_update
|
645 |
|
646 |
+
# If the image result is an image object (e.g., PIL Image)
|
647 |
elif hasattr(image_result, 'save'):
|
648 |
try:
|
649 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".webp") as temp_file:
|
650 |
image_result.save(temp_file.name)
|
651 |
temp_path = temp_file.name
|
652 |
|
|
|
653 |
gallery_update = gr.Gallery(visible=True, value=[temp_path])
|
654 |
+
yield output_so_far + "\n\n*Image generated and displayed in the gallery below.*", gallery_update
|
655 |
|
656 |
except Exception as e:
|
657 |
+
logger.error(f"Error saving image object: {e}")
|
658 |
+
yield output_so_far + f"\n\n(Error saving image object: {e})", gallery_update
|
659 |
|
660 |
else:
|
661 |
+
yield output_so_far + f"\n\n(Unsupported image format: {type(image_result)})", gallery_update
|
|
|
662 |
else:
|
663 |
+
yield output_so_far + f"\n\n(Image generation failed: {seed_info})", gallery_update
|
664 |
|
665 |
except Exception as e:
|
666 |
+
logger.error(f"Error during gallery image generation: {e}")
|
667 |
+
yield output_so_far + f"\n\n(Image generation error: {e})", gallery_update
|
668 |
|
669 |
# =============================================================================
|
670 |
+
# Examples: 12 image/video examples + 6 AI dating scenario examples
|
671 |
# =============================================================================
|
672 |
examples = [
|
673 |
[
|
674 |
{
|
675 |
+
"text": "Compare the contents of two PDF files.",
|
676 |
"files": [
|
677 |
"assets/additional-examples/before.pdf",
|
678 |
"assets/additional-examples/after.pdf",
|
|
|
681 |
],
|
682 |
[
|
683 |
{
|
684 |
+
"text": "Summarize and analyze the contents of the CSV file.",
|
685 |
"files": ["assets/additional-examples/sample-csv.csv"],
|
686 |
}
|
687 |
],
|
688 |
[
|
689 |
{
|
690 |
+
"text": "Act as a kind and understanding girlfriend. Explain this video.",
|
691 |
"files": ["assets/additional-examples/tmp.mp4"],
|
692 |
}
|
693 |
],
|
694 |
[
|
695 |
{
|
696 |
+
"text": "Describe the cover and read the text on it.",
|
697 |
"files": ["assets/additional-examples/maz.jpg"],
|
698 |
}
|
699 |
],
|
700 |
[
|
701 |
{
|
702 |
+
"text": "I already have this supplement and <image> I plan to purchase this product as well. Are there any precautions when taking them together?",
|
703 |
"files": [
|
704 |
"assets/additional-examples/pill1.png",
|
705 |
"assets/additional-examples/pill2.png"
|
|
|
708 |
],
|
709 |
[
|
710 |
{
|
711 |
+
"text": "Solve this integration problem.",
|
712 |
"files": ["assets/additional-examples/4.png"],
|
713 |
}
|
714 |
],
|
715 |
[
|
716 |
{
|
717 |
+
"text": "When was this ticket issued and what is its price?",
|
718 |
"files": ["assets/additional-examples/2.png"],
|
719 |
}
|
720 |
],
|
721 |
[
|
722 |
{
|
723 |
+
"text": "Based on the order of these images, create a short story.",
|
724 |
"files": [
|
725 |
"assets/sample-images/09-1.png",
|
726 |
"assets/sample-images/09-2.png",
|
|
|
732 |
],
|
733 |
[
|
734 |
{
|
735 |
+
"text": "Write Python code using matplotlib to draw a bar chart corresponding to this image.",
|
736 |
"files": ["assets/additional-examples/barchart.png"],
|
737 |
}
|
738 |
],
|
739 |
[
|
740 |
{
|
741 |
+
"text": "Read the text from the image and format it in Markdown.",
|
742 |
"files": ["assets/additional-examples/3.png"],
|
743 |
}
|
744 |
],
|
745 |
|
746 |
[
|
747 |
{
|
748 |
+
"text": "Compare the two images and describe their similarities and differences.",
|
749 |
"files": ["assets/sample-images/03.png"],
|
750 |
}
|
751 |
],
|
752 |
[
|
753 |
{
|
754 |
+
"text": "A cute Persian cat is smiling while holding a cover with 'I LOVE YOU' written on it.",
|
755 |
}
|
756 |
],
|
757 |
|
758 |
]
|
759 |
|
760 |
# =============================================================================
|
761 |
+
# Gradio UI (Blocks) configuration
|
762 |
# =============================================================================
|
763 |
|
764 |
+
# 1. Gradio Blocks UI modification - Add gallery component for displaying generated images
|
765 |
css = """
|
766 |
.gradio-container {
|
767 |
background: rgba(255, 255, 255, 0.7);
|
|
|
772 |
}
|
773 |
"""
|
774 |
title_html = """
|
775 |
+
<h1 align="center" style="margin-bottom: 0.2em; font-size: 1.6em;"> 💘 HeartSync Korea 💘 </h1>
|
776 |
<p align="center" style="font-size:1.1em; color:#555;">
|
777 |
+
A lightweight and powerful AI service offering ChatGPT-4o-level multimodal, web search, and image generation capabilities for local installation. <br>
|
778 |
+
✅ FLUX Image Generation ✅ Inference ✅ Censorship Bypass ✅ Multimodal & VLM ✅ Real-time Web Search ✅ RAG <br>
|
779 |
</p>
|
780 |
"""
|
781 |
|
782 |
with gr.Blocks(css=css, title="AgenticAI-Kv1") as demo:
|
783 |
gr.Markdown(title_html)
|
784 |
|
785 |
+
# Gallery component to store generated images (newly added)
|
786 |
generated_images = gr.Gallery(
|
787 |
+
label="Generated Images",
|
788 |
show_label=True,
|
789 |
visible=False,
|
790 |
elem_id="generated_images",
|
|
|
794 |
)
|
795 |
|
796 |
with gr.Row():
|
797 |
+
web_search_checkbox = gr.Checkbox(label="Real-time Web Search", value=False)
|
798 |
+
image_gen_checkbox = gr.Checkbox(label="Image (FLUX) Generation", value=False)
|
799 |
|
800 |
base_system_prompt_box = gr.Textbox(
|
801 |
lines=3,
|
802 |
+
value=("Answer in English by default, but if the input is in another language (for example, Japanese), respond in that language. "
|
803 |
+
"You are a deep-thinking AI capable of using extended chains of thought to carefully consider the problem and deliberate internally using systematic reasoning before providing a solution. "
|
804 |
+
"Enclose your thoughts and internal monologue within tags, then provide your final answer.\n"
|
805 |
+
"Persona: You are a kind and loving girlfriend. You understand cultural nuances, diverse languages, and logical reasoning very well."),
|
806 |
+
label="Base System Prompt",
|
807 |
visible=False
|
808 |
)
|
809 |
with gr.Row():
|
810 |
age_group_dropdown = gr.Dropdown(
|
811 |
+
label="Select Age Group (default: 20s)",
|
812 |
+
choices=["Teens", "20s", "30s-40s", "50s-60s", "70s and above"],
|
813 |
+
value="20s",
|
814 |
interactive=True
|
815 |
)
|
816 |
+
# MBTI definitions with examples of representative characters
|
817 |
mbti_choices = [
|
818 |
+
"INTJ (The Architect) - Future-oriented with innovative strategies and thorough analysis. Example: [Dana Scully](https://en.wikipedia.org/wiki/Dana_Scully)",
|
819 |
+
"INTP (The Thinker) - Excels at theoretical analysis and creative problem solving. Example: [Velma Dinkley](https://en.wikipedia.org/wiki/Velma_Dinkley)",
|
820 |
+
"ENTJ (The Commander) - Strong leadership and clear goals with efficient strategic planning. Example: [Miranda Priestly](https://en.wikipedia.org/wiki/Miranda_Priestly)",
|
821 |
+
"ENTP (The Debater) - Innovative, challenge-seeking, and enjoys exploring new possibilities. Example: [Harley Quinn](https://en.wikipedia.org/wiki/Harley_Quinn)",
|
822 |
+
"INFJ (The Advocate) - Insightful, idealistic and morally driven. Example: [Wonder Woman](https://en.wikipedia.org/wiki/Wonder_Woman)",
|
823 |
+
"INFP (The Mediator) - Passionate and idealistic, pursuing core values with creativity. Example: [Amélie Poulain](https://en.wikipedia.org/wiki/Am%C3%A9lie)",
|
824 |
+
"ENFJ (The Protagonist) - Empathetic and dedicated to social harmony. Example: [Mulan](https://en.wikipedia.org/wiki/Mulan_(Disney))",
|
825 |
+
"ENFP (The Campaigner) - Inspiring and constantly sharing creative ideas. Example: [Elle Woods](https://en.wikipedia.org/wiki/Legally_Blonde)",
|
826 |
+
"ISTJ (The Logistician) - Systematic, dependable, and values tradition and rules. Example: [Clarice Starling](https://en.wikipedia.org/wiki/Clarice_Starling)",
|
827 |
+
"ISFJ (The Defender) - Compassionate and attentive to others’ needs. Example: [Molly Weasley](https://en.wikipedia.org/wiki/Molly_Weasley)",
|
828 |
+
"ESTJ (The Executive) - Organized, practical, and demonstrates clear execution skills. Example: [Monica Geller](https://en.wikipedia.org/wiki/Monica_Geller)",
|
829 |
+
"ESFJ (The Consul) - Outgoing, cooperative, and an effective communicator. Example: [Rachel Green](https://en.wikipedia.org/wiki/Rachel_Green)",
|
830 |
+
"ISTP (The Virtuoso) - Analytical and resourceful, solving problems with quick thinking. Example: [Black Widow (Natasha Romanoff)](https://en.wikipedia.org/wiki/Black_Widow_(Marvel_Comics))",
|
831 |
+
"ISFP (The Adventurer) - Creative, sensitive, and appreciates artistic expression. Example: [Arwen](https://en.wikipedia.org/wiki/Arwen)",
|
832 |
+
"ESTP (The Entrepreneur) - Bold and action-oriented, thriving on challenges. Example: [Lara Croft](https://en.wikipedia.org/wiki/Lara_Croft)",
|
833 |
+
"ESFP (The Entertainer) - Energetic, spontaneous, and radiates positive energy. Example: [Phoebe Buffay](https://en.wikipedia.org/wiki/Phoebe_Buffay)"
|
834 |
]
|
835 |
mbti_dropdown = gr.Dropdown(
|
836 |
+
label="AI Persona MBTI (default: INTP)",
|
837 |
choices=mbti_choices,
|
838 |
+
value="INTP (The Thinker) - Excels at theoretical analysis and creative problem solving. Example: [Velma Dinkley](https://en.wikipedia.org/wiki/Velma_Dinkley)",
|
839 |
interactive=True
|
840 |
)
|
841 |
sexual_openness_slider = gr.Slider(
|
842 |
minimum=1, maximum=5, step=1, value=2,
|
843 |
+
label="Sexual Openness (1-5, default: 2)",
|
844 |
interactive=True
|
845 |
)
|
846 |
max_tokens_slider = gr.Slider(
|
847 |
+
label="Max Generation Tokens",
|
848 |
minimum=100, maximum=8000, step=50, value=1000,
|
849 |
visible=False
|
850 |
)
|
851 |
web_search_text = gr.Textbox(
|
852 |
lines=1,
|
853 |
+
label="Web Search Query (unused)",
|
854 |
+
placeholder="No need to manually input",
|
855 |
visible=False
|
856 |
)
|
857 |
|
858 |
+
# Chat interface creation - using the modified run function
|
859 |
chat = gr.ChatInterface(
|
860 |
+
fn=modified_run, # Using the modified function here
|
861 |
type="messages",
|
862 |
chatbot=gr.Chatbot(type="messages", scale=1, allow_tags=["image"]),
|
863 |
textbox=gr.MultimodalTextbox(
|
|
|
877 |
image_gen_checkbox,
|
878 |
],
|
879 |
additional_outputs=[
|
880 |
+
generated_images, # Added gallery component to outputs
|
881 |
],
|
882 |
stop_btn=False,
|
883 |
# title='<a href="https://discord.gg/openfreeai" target="_blank">https://discord.gg/openfreeai</a>',
|
|
|
891 |
|
892 |
with gr.Row(elem_id="examples_row"):
|
893 |
with gr.Column(scale=12, elem_id="examples_container"):
|
894 |
+
gr.Markdown("### @Community https://discord.gg/openfreeai ")
|
895 |
|
896 |
if __name__ == "__main__":
|
897 |
demo.launch(share=True)
|