Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -14,30 +14,26 @@ from loguru import logger
|
|
14 |
from PIL import Image
|
15 |
from transformers import AutoProcessor, Gemma3ForConditionalGeneration, TextIteratorStreamer
|
16 |
|
17 |
-
#
|
18 |
import pandas as pd
|
19 |
|
20 |
-
|
21 |
-
# ์ ์ฒด ์ ๋ฌธ์ ๋๊ธฐ๋, ๋๋ฌด ํด ๊ฒฝ์ฐ ์๋ผ๋ด๊ธฐ ์ํ ์์
|
22 |
-
##################################################
|
23 |
-
MAX_CONTENT_CHARS = 8000 # ์: 8000์ ์ด๊ณผ ์ ์๋ผ๋
|
24 |
|
25 |
model_id = os.getenv("MODEL_ID", "google/gemma-3-27b-it")
|
26 |
processor = AutoProcessor.from_pretrained(model_id, padding_side="left")
|
27 |
model = Gemma3ForConditionalGeneration.from_pretrained(
|
28 |
-
model_id,
|
|
|
|
|
|
|
29 |
)
|
30 |
|
31 |
MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "5"))
|
32 |
|
33 |
|
34 |
-
##################################################
|
35 |
-
# CSV/TXT ์ ๋ฌธ ์ฒ๋ฆฌ ํจ์
|
36 |
-
##################################################
|
37 |
def analyze_csv_file(path: str) -> str:
|
38 |
"""
|
39 |
-
CSV
|
40 |
-
๋๋ฌด ๊ธธ๋ฉด MAX_CONTENT_CHARS๊น์ง๋ง ์๋ผ๋.
|
41 |
"""
|
42 |
try:
|
43 |
df = pd.read_csv(path)
|
@@ -45,37 +41,26 @@ def analyze_csv_file(path: str) -> str:
|
|
45 |
if len(df_str) > MAX_CONTENT_CHARS:
|
46 |
df_str = df_str[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
47 |
|
48 |
-
return (
|
49 |
-
f"**[CSV File: {os.path.basename(path)}]**\n\n"
|
50 |
-
f"{df_str}"
|
51 |
-
)
|
52 |
except Exception as e:
|
53 |
return f"Failed to read CSV ({os.path.basename(path)}): {str(e)}"
|
54 |
|
55 |
|
56 |
def analyze_txt_file(path: str) -> str:
|
57 |
"""
|
58 |
-
TXT ํ์ผ
|
59 |
-
๋๋ฌด ๊ธธ๋ฉด MAX_CONTENT_CHARS๊น์ง๋ง ์๋ผ๋.
|
60 |
"""
|
61 |
try:
|
62 |
with open(path, "r", encoding="utf-8") as f:
|
63 |
text = f.read()
|
64 |
-
|
65 |
if len(text) > MAX_CONTENT_CHARS:
|
66 |
text = text[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
67 |
|
68 |
-
return (
|
69 |
-
f"**[TXT File: {os.path.basename(path)}]**\n\n"
|
70 |
-
f"{text}"
|
71 |
-
)
|
72 |
except Exception as e:
|
73 |
return f"Failed to read TXT ({os.path.basename(path)}): {str(e)}"
|
74 |
|
75 |
|
76 |
-
##################################################
|
77 |
-
# ๊ธฐ์กด ๋ฏธ๋์ด ํ์ผ ๊ฒ์ฌ ๋ก์ง (์ด๋ฏธ์ง/๋น๋์ค)
|
78 |
-
##################################################
|
79 |
def count_files_in_new_message(paths: list[str]) -> tuple[int, int]:
|
80 |
image_count = 0
|
81 |
video_count = 0
|
@@ -105,14 +90,13 @@ def validate_media_constraints(message: dict, history: list[dict]) -> bool:
|
|
105 |
- ๋น๋์ค 1๊ฐ ์ด๊ณผ ๋ถ๊ฐ
|
106 |
- ๋น๋์ค/์ด๋ฏธ์ง ํผํฉ ๋ถ๊ฐ
|
107 |
- ์ด๋ฏธ์ง ๊ฐ์ MAX_NUM_IMAGES ์ด๊ณผ ๋ถ๊ฐ
|
108 |
-
- <image> ํ๊ทธ๊ฐ ์์ผ๋ฉด ํ๊ทธ ์์ ์ด๋ฏธ์ง
|
109 |
-
CSV, TXT, PDF ๋ฑ์ ์ฌ๊ธฐ์ ์ ํํ์ง ์์.
|
110 |
"""
|
111 |
media_files = []
|
112 |
for f in message["files"]:
|
113 |
-
#
|
114 |
-
|
115 |
-
if f.endswith(".mp4") or re.search(r"\.(png|jpg|jpeg|gif|webp)$", f, re.IGNORECASE):
|
116 |
media_files.append(f)
|
117 |
|
118 |
new_image_count, new_video_count = count_files_in_new_message(media_files)
|
@@ -140,9 +124,6 @@ def validate_media_constraints(message: dict, history: list[dict]) -> bool:
|
|
140 |
return True
|
141 |
|
142 |
|
143 |
-
##################################################
|
144 |
-
# ๋น๋์ค ์ฒ๋ฆฌ
|
145 |
-
##################################################
|
146 |
def downsample_video(video_path: str) -> list[tuple[Image.Image, float]]:
|
147 |
vidcap = cv2.VideoCapture(video_path)
|
148 |
fps = vidcap.get(cv2.CAP_PROP_FPS)
|
@@ -177,9 +158,6 @@ def process_video(video_path: str) -> list[dict]:
|
|
177 |
return content
|
178 |
|
179 |
|
180 |
-
##################################################
|
181 |
-
# interleaved <image> ํ๊ทธ ์ฒ๋ฆฌ
|
182 |
-
##################################################
|
183 |
def process_interleaved_images(message: dict) -> list[dict]:
|
184 |
logger.debug(f"{message['files']=}")
|
185 |
parts = re.split(r"(<image>)", message["text"])
|
@@ -188,7 +166,6 @@ def process_interleaved_images(message: dict) -> list[dict]:
|
|
188 |
content = []
|
189 |
image_index = 0
|
190 |
for part in parts:
|
191 |
-
logger.debug(f"{part=}")
|
192 |
if part == "<image>":
|
193 |
content.append({"type": "image", "url": message["files"][image_index]})
|
194 |
logger.debug(f"file: {message['files'][image_index]}")
|
@@ -201,16 +178,7 @@ def process_interleaved_images(message: dict) -> list[dict]:
|
|
201 |
return content
|
202 |
|
203 |
|
204 |
-
##################################################
|
205 |
-
# CSV, TXT ํ์ผ๋ ์ ๋ฌธ์ LLM์ ๋๊ธฐ๋๋ก
|
206 |
-
##################################################
|
207 |
def process_new_user_message(message: dict) -> list[dict]:
|
208 |
-
"""
|
209 |
-
- mp4 -> ๋น๋์ค ์ฒ๋ฆฌ
|
210 |
-
- ์ด๋ฏธ์ง -> interleaved or multiple
|
211 |
-
- CSV -> ์ ์ฒด df.to_string() (๋๋ฌด ๊ธธ๋ฉด ์๋ผ๋)
|
212 |
-
- TXT -> ์ ์ฒด text (๋๋ฌด ๊ธธ๋ฉด ์๋ผ๋)
|
213 |
-
"""
|
214 |
if not message["files"]:
|
215 |
return [{"type": "text", "text": message["text"]}]
|
216 |
|
@@ -220,7 +188,7 @@ def process_new_user_message(message: dict) -> list[dict]:
|
|
220 |
csv_files = [f for f in message["files"] if f.lower().endswith(".csv")]
|
221 |
txt_files = [f for f in message["files"] if f.lower().endswith(".txt")]
|
222 |
|
223 |
-
# ์ฌ์ฉ์
|
224 |
content_list = [{"type": "text", "text": message["text"]}]
|
225 |
|
226 |
# CSV ์ ๋ฌธ
|
@@ -233,7 +201,7 @@ def process_new_user_message(message: dict) -> list[dict]:
|
|
233 |
txt_analysis = analyze_txt_file(txt_path)
|
234 |
content_list.append({"type": "text", "text": txt_analysis})
|
235 |
|
236 |
-
#
|
237 |
if video_files:
|
238 |
content_list += process_video(video_files[0])
|
239 |
return content_list
|
@@ -242,7 +210,7 @@ def process_new_user_message(message: dict) -> list[dict]:
|
|
242 |
if "<image>" in message["text"]:
|
243 |
return process_interleaved_images(message)
|
244 |
|
245 |
-
# ์ผ๋ฐ
|
246 |
if image_files:
|
247 |
for img_path in image_files:
|
248 |
content_list.append({"type": "image", "url": img_path})
|
@@ -250,9 +218,6 @@ def process_new_user_message(message: dict) -> list[dict]:
|
|
250 |
return content_list
|
251 |
|
252 |
|
253 |
-
##################################################
|
254 |
-
# history -> LLM ๋ฉ์์ง ๋ณํ
|
255 |
-
##################################################
|
256 |
def process_history(history: list[dict]) -> list[dict]:
|
257 |
messages = []
|
258 |
current_user_content: list[dict] = []
|
@@ -271,9 +236,6 @@ def process_history(history: list[dict]) -> list[dict]:
|
|
271 |
return messages
|
272 |
|
273 |
|
274 |
-
##################################################
|
275 |
-
# ๋ฉ์ธ ์ถ๋ก ํจ์
|
276 |
-
##################################################
|
277 |
@spaces.GPU(duration=120)
|
278 |
def run(message: dict, history: list[dict], system_prompt: str = "", max_new_tokens: int = 512) -> Iterator[str]:
|
279 |
if not validate_media_constraints(message, history):
|
@@ -309,9 +271,6 @@ def run(message: dict, history: list[dict], system_prompt: str = "", max_new_tok
|
|
309 |
yield output
|
310 |
|
311 |
|
312 |
-
##################################################
|
313 |
-
# ์์ ๋ชฉ๋ก (๊ธฐ์กด)
|
314 |
-
##################################################
|
315 |
examples = [
|
316 |
[
|
317 |
{
|
@@ -435,16 +394,16 @@ examples = [
|
|
435 |
]
|
436 |
|
437 |
|
438 |
-
##################################################
|
439 |
-
# Gradio ChatInterface
|
440 |
-
##################################################
|
441 |
demo = gr.ChatInterface(
|
442 |
fn=run,
|
443 |
type="messages",
|
444 |
chatbot=gr.Chatbot(type="messages", scale=1, allow_tags=["image"]),
|
445 |
-
#
|
446 |
textbox=gr.MultimodalTextbox(
|
447 |
-
file_types=[
|
|
|
|
|
|
|
448 |
file_count="multiple",
|
449 |
autofocus=True
|
450 |
),
|
@@ -452,18 +411,9 @@ demo = gr.ChatInterface(
|
|
452 |
additional_inputs=[
|
453 |
gr.Textbox(
|
454 |
label="System Prompt",
|
455 |
-
value=
|
456 |
-
"You are a deeply thoughtful AI. Consider problems thoroughly and derive "
|
457 |
-
"correct solutions through systematic reasoning. Please answer in korean."
|
458 |
-
)
|
459 |
-
),
|
460 |
-
gr.Slider(
|
461 |
-
label="Max New Tokens",
|
462 |
-
minimum=100,
|
463 |
-
maximum=8000,
|
464 |
-
step=50,
|
465 |
-
value=2000
|
466 |
),
|
|
|
467 |
],
|
468 |
stop_btn=False,
|
469 |
title="Gemma 3 27B IT",
|
|
|
14 |
from PIL import Image
|
15 |
from transformers import AutoProcessor, Gemma3ForConditionalGeneration, TextIteratorStreamer
|
16 |
|
17 |
+
# CSV/TXT ๋ถ์
|
18 |
import pandas as pd
|
19 |
|
20 |
+
MAX_CONTENT_CHARS = 8000 # ํ์ผ์์ ์ฝ์ ๋ด์ฉ์ด ๋๋ฌด ๊ธธ ๊ฒฝ์ฐ ์ด ์ ๋์์ ์๋ผ๋
|
|
|
|
|
|
|
21 |
|
22 |
model_id = os.getenv("MODEL_ID", "google/gemma-3-27b-it")
|
23 |
processor = AutoProcessor.from_pretrained(model_id, padding_side="left")
|
24 |
model = Gemma3ForConditionalGeneration.from_pretrained(
|
25 |
+
model_id,
|
26 |
+
device_map="auto",
|
27 |
+
torch_dtype=torch.bfloat16,
|
28 |
+
attn_implementation="eager"
|
29 |
)
|
30 |
|
31 |
MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "5"))
|
32 |
|
33 |
|
|
|
|
|
|
|
34 |
def analyze_csv_file(path: str) -> str:
|
35 |
"""
|
36 |
+
CSV ํ์ผ์ ์ฝ์ด ๋ฌธ์์ดํ. ๋๋ฌด ํฌ๋ฉด ์ผ๋ถ๋ง ์๋ผ๋.
|
|
|
37 |
"""
|
38 |
try:
|
39 |
df = pd.read_csv(path)
|
|
|
41 |
if len(df_str) > MAX_CONTENT_CHARS:
|
42 |
df_str = df_str[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
43 |
|
44 |
+
return f"**[CSV File: {os.path.basename(path)}]**\n\n{df_str}"
|
|
|
|
|
|
|
45 |
except Exception as e:
|
46 |
return f"Failed to read CSV ({os.path.basename(path)}): {str(e)}"
|
47 |
|
48 |
|
49 |
def analyze_txt_file(path: str) -> str:
|
50 |
"""
|
51 |
+
TXT ํ์ผ ์ ๋ฌธ ์ฝ์ด๋ค์ด๋, ๋๋ฌด ๊ธธ๋ฉด ์๋ผ๋.
|
|
|
52 |
"""
|
53 |
try:
|
54 |
with open(path, "r", encoding="utf-8") as f:
|
55 |
text = f.read()
|
|
|
56 |
if len(text) > MAX_CONTENT_CHARS:
|
57 |
text = text[:MAX_CONTENT_CHARS] + "\n...(truncated)..."
|
58 |
|
59 |
+
return f"**[TXT File: {os.path.basename(path)}]**\n\n{text}"
|
|
|
|
|
|
|
60 |
except Exception as e:
|
61 |
return f"Failed to read TXT ({os.path.basename(path)}): {str(e)}"
|
62 |
|
63 |
|
|
|
|
|
|
|
64 |
def count_files_in_new_message(paths: list[str]) -> tuple[int, int]:
|
65 |
image_count = 0
|
66 |
video_count = 0
|
|
|
90 |
- ๋น๋์ค 1๊ฐ ์ด๊ณผ ๋ถ๊ฐ
|
91 |
- ๋น๋์ค/์ด๋ฏธ์ง ํผํฉ ๋ถ๊ฐ
|
92 |
- ์ด๋ฏธ์ง ๊ฐ์ MAX_NUM_IMAGES ์ด๊ณผ ๋ถ๊ฐ
|
93 |
+
- <image> ํ๊ทธ๊ฐ ์์ผ๋ฉด ํ๊ทธ ์์ ์ค์ ์ด๋ฏธ์ง ๊ฐ์ ์ผ์น
|
94 |
+
- CSV, TXT, PDF ๋ฑ์ ์ฌ๊ธฐ์ ์ ํํ์ง ์์.
|
95 |
"""
|
96 |
media_files = []
|
97 |
for f in message["files"]:
|
98 |
+
# ์ด๋ฏธ์ง(์ฌ๋ฌ ํ์ฅ์)๋ mp4๋ง ์ฒดํฌ
|
99 |
+
if re.search(r"\.(png|jpg|jpeg|gif|webp)$", f, re.IGNORECASE) or f.endswith(".mp4"):
|
|
|
100 |
media_files.append(f)
|
101 |
|
102 |
new_image_count, new_video_count = count_files_in_new_message(media_files)
|
|
|
124 |
return True
|
125 |
|
126 |
|
|
|
|
|
|
|
127 |
def downsample_video(video_path: str) -> list[tuple[Image.Image, float]]:
|
128 |
vidcap = cv2.VideoCapture(video_path)
|
129 |
fps = vidcap.get(cv2.CAP_PROP_FPS)
|
|
|
158 |
return content
|
159 |
|
160 |
|
|
|
|
|
|
|
161 |
def process_interleaved_images(message: dict) -> list[dict]:
|
162 |
logger.debug(f"{message['files']=}")
|
163 |
parts = re.split(r"(<image>)", message["text"])
|
|
|
166 |
content = []
|
167 |
image_index = 0
|
168 |
for part in parts:
|
|
|
169 |
if part == "<image>":
|
170 |
content.append({"type": "image", "url": message["files"][image_index]})
|
171 |
logger.debug(f"file: {message['files'][image_index]}")
|
|
|
178 |
return content
|
179 |
|
180 |
|
|
|
|
|
|
|
181 |
def process_new_user_message(message: dict) -> list[dict]:
|
|
|
|
|
|
|
|
|
|
|
|
|
182 |
if not message["files"]:
|
183 |
return [{"type": "text", "text": message["text"]}]
|
184 |
|
|
|
188 |
csv_files = [f for f in message["files"] if f.lower().endswith(".csv")]
|
189 |
txt_files = [f for f in message["files"] if f.lower().endswith(".txt")]
|
190 |
|
191 |
+
# ์ฌ์ฉ์ ์
๋ ฅ ํ
์คํธ๋ฅผ ๋จผ์
|
192 |
content_list = [{"type": "text", "text": message["text"]}]
|
193 |
|
194 |
# CSV ์ ๋ฌธ
|
|
|
201 |
txt_analysis = analyze_txt_file(txt_path)
|
202 |
content_list.append({"type": "text", "text": txt_analysis})
|
203 |
|
204 |
+
# ๋์์ ์ฒ๋ฆฌ
|
205 |
if video_files:
|
206 |
content_list += process_video(video_files[0])
|
207 |
return content_list
|
|
|
210 |
if "<image>" in message["text"]:
|
211 |
return process_interleaved_images(message)
|
212 |
|
213 |
+
# ์ผ๋ฐ ์ด๋ฏธ์ง๋ค
|
214 |
if image_files:
|
215 |
for img_path in image_files:
|
216 |
content_list.append({"type": "image", "url": img_path})
|
|
|
218 |
return content_list
|
219 |
|
220 |
|
|
|
|
|
|
|
221 |
def process_history(history: list[dict]) -> list[dict]:
|
222 |
messages = []
|
223 |
current_user_content: list[dict] = []
|
|
|
236 |
return messages
|
237 |
|
238 |
|
|
|
|
|
|
|
239 |
@spaces.GPU(duration=120)
|
240 |
def run(message: dict, history: list[dict], system_prompt: str = "", max_new_tokens: int = 512) -> Iterator[str]:
|
241 |
if not validate_media_constraints(message, history):
|
|
|
271 |
yield output
|
272 |
|
273 |
|
|
|
|
|
|
|
274 |
examples = [
|
275 |
[
|
276 |
{
|
|
|
394 |
]
|
395 |
|
396 |
|
|
|
|
|
|
|
397 |
demo = gr.ChatInterface(
|
398 |
fn=run,
|
399 |
type="messages",
|
400 |
chatbot=gr.Chatbot(type="messages", scale=1, allow_tags=["image"]),
|
401 |
+
# .webp, .png, .jpg, .jpeg, .gif, .mp4, .csv, .txt, .pdf ๋ชจ๋ ํ์ฉ
|
402 |
textbox=gr.MultimodalTextbox(
|
403 |
+
file_types=[
|
404 |
+
".webp", ".png", ".jpg", ".jpeg", ".gif",
|
405 |
+
".mp4", ".csv", ".txt", ".pdf"
|
406 |
+
],
|
407 |
file_count="multiple",
|
408 |
autofocus=True
|
409 |
),
|
|
|
411 |
additional_inputs=[
|
412 |
gr.Textbox(
|
413 |
label="System Prompt",
|
414 |
+
value="You are a deeply thoughtful AI. Consider problems thoroughly and derive correct solutions through systematic reasoning. Please answer in korean."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
415 |
),
|
416 |
+
gr.Slider(label="Max New Tokens", minimum=100, maximum=8000, step=50, value=2000),
|
417 |
],
|
418 |
stop_btn=False,
|
419 |
title="Gemma 3 27B IT",
|