cutechicken commited on
Commit
2317674
β€’
1 Parent(s): ebbebcc

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +315 -0
app.py ADDED
@@ -0,0 +1,315 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ from dotenv import load_dotenv
3
+ import gradio as gr
4
+ from huggingface_hub import InferenceClient
5
+ import pandas as pd
6
+ from typing import List, Tuple
7
+ import json
8
+ from datetime import datetime
9
+
10
+ # ν™˜κ²½ λ³€μˆ˜ μ„€μ •
11
+ HF_TOKEN = os.getenv("HF_TOKEN")
12
+
13
+ # LLM Models Definition
14
+ LLM_MODELS = {
15
+ "Cohere c4ai-crp-08-2024": "CohereForAI/c4ai-command-r-plus-08-2024", # Default
16
+ "Meta Llama3.3-70B": "meta-llama/Llama-3.3-70B-Instruct" # Backup model
17
+ }
18
+
19
+ class ChatHistory:
20
+ def __init__(self):
21
+ self.history = []
22
+ self.history_file = "/tmp/chat_history.json"
23
+ self.load_history()
24
+
25
+ def add_conversation(self, user_msg: str, assistant_msg: str):
26
+ conversation = {
27
+ "timestamp": datetime.now().isoformat(),
28
+ "messages": [
29
+ {"role": "user", "content": user_msg},
30
+ {"role": "assistant", "content": assistant_msg}
31
+ ]
32
+ }
33
+ self.history.append(conversation)
34
+ self.save_history()
35
+
36
+ def format_for_display(self):
37
+ # Gradio Chatbot μ»΄ν¬λ„ŒνŠΈμ— λ§žλŠ” ν˜•μ‹μœΌλ‘œ λ³€ν™˜
38
+ formatted = []
39
+ for conv in self.history:
40
+ formatted.append([
41
+ conv["messages"][0]["content"], # user message
42
+ conv["messages"][1]["content"] # assistant message
43
+ ])
44
+ return formatted
45
+
46
+ def get_messages_for_api(self):
47
+ # API ν˜ΈμΆœμ„ μœ„ν•œ λ©”μ‹œμ§€ ν˜•μ‹
48
+ messages = []
49
+ for conv in self.history:
50
+ messages.extend([
51
+ {"role": "user", "content": conv["messages"][0]["content"]},
52
+ {"role": "assistant", "content": conv["messages"][1]["content"]}
53
+ ])
54
+ return messages
55
+
56
+ def clear_history(self):
57
+ self.history = []
58
+ self.save_history()
59
+
60
+ def save_history(self):
61
+ try:
62
+ with open(self.history_file, 'w', encoding='utf-8') as f:
63
+ json.dump(self.history, f, ensure_ascii=False, indent=2)
64
+ except Exception as e:
65
+ print(f"νžˆμŠ€ν† λ¦¬ μ €μž₯ μ‹€νŒ¨: {e}")
66
+
67
+ def load_history(self):
68
+ try:
69
+ if os.path.exists(self.history_file):
70
+ with open(self.history_file, 'r', encoding='utf-8') as f:
71
+ self.history = json.load(f)
72
+ except Exception as e:
73
+ print(f"νžˆμŠ€ν† λ¦¬ λ‘œλ“œ μ‹€νŒ¨: {e}")
74
+ self.history = []
75
+
76
+
77
+ # μ „μ—­ ChatHistory μΈμŠ€ν„΄μŠ€ 생성
78
+ chat_history = ChatHistory()
79
+
80
+ def get_client(model_name="Cohere c4ai-crp-08-2024"):
81
+ try:
82
+ return InferenceClient(LLM_MODELS[model_name], token=HF_TOKEN)
83
+ except Exception:
84
+ return InferenceClient(LLM_MODELS["Meta Llama3.3-70B"], token=HF_TOKEN)
85
+
86
+ def analyze_file_content(content, file_type):
87
+ """Analyze file content and return structural summary"""
88
+ if file_type in ['parquet', 'csv']:
89
+ try:
90
+ lines = content.split('\n')
91
+ header = lines[0]
92
+ columns = header.count('|') - 1
93
+ rows = len(lines) - 3
94
+ return f"πŸ“Š 데이터셋 ꡬ쑰: {columns}개 컬럼, {rows}개 데이터"
95
+ except:
96
+ return "❌ 데이터셋 ꡬ쑰 뢄석 μ‹€νŒ¨"
97
+
98
+ lines = content.split('\n')
99
+ total_lines = len(lines)
100
+ non_empty_lines = len([line for line in lines if line.strip()])
101
+
102
+ if any(keyword in content.lower() for keyword in ['def ', 'class ', 'import ', 'function']):
103
+ functions = len([line for line in lines if 'def ' in line])
104
+ classes = len([line for line in lines if 'class ' in line])
105
+ imports = len([line for line in lines if 'import ' in line or 'from ' in line])
106
+ return f"πŸ’» μ½”λ“œ ꡬ쑰: {total_lines}쀄 (ν•¨μˆ˜: {functions}, 클래슀: {classes}, μž„ν¬νŠΈ: {imports})"
107
+
108
+ paragraphs = content.count('\n\n') + 1
109
+ words = len(content.split())
110
+ return f"πŸ“ λ¬Έμ„œ ꡬ쑰: {total_lines}쀄, {paragraphs}단락, μ•½ {words}단어"
111
+
112
+ def read_uploaded_file(file):
113
+ if file is None:
114
+ return "", ""
115
+ try:
116
+ file_ext = os.path.splitext(file.name)[1].lower()
117
+
118
+ if file_ext == '.parquet':
119
+ df = pd.read_parquet(file.name, engine='pyarrow')
120
+ content = df.head(10).to_markdown(index=False)
121
+ return content, "parquet"
122
+ elif file_ext == '.csv':
123
+ encodings = ['utf-8', 'cp949', 'euc-kr', 'latin1']
124
+ for encoding in encodings:
125
+ try:
126
+ df = pd.read_csv(file.name, encoding=encoding)
127
+ content = f"πŸ“Š 데이터 미리보기:\n{df.head(10).to_markdown(index=False)}\n\n"
128
+ content += f"\nπŸ“ˆ 데이터 정보:\n"
129
+ content += f"- 전체 ν–‰ 수: {len(df)}\n"
130
+ content += f"- 전체 μ—΄ 수: {len(df.columns)}\n"
131
+ content += f"- 컬럼 λͺ©λ‘: {', '.join(df.columns)}\n"
132
+ content += f"\nπŸ“‹ 컬럼 데이터 νƒ€μž…:\n"
133
+ for col, dtype in df.dtypes.items():
134
+ content += f"- {col}: {dtype}\n"
135
+ null_counts = df.isnull().sum()
136
+ if null_counts.any():
137
+ content += f"\n⚠️ 결츑치:\n"
138
+ for col, null_count in null_counts[null_counts > 0].items():
139
+ content += f"- {col}: {null_count}개 λˆ„λ½\n"
140
+ return content, "csv"
141
+ except UnicodeDecodeError:
142
+ continue
143
+ raise UnicodeDecodeError(f"❌ μ§€μ›λ˜λŠ” μΈμ½”λ”©μœΌλ‘œ νŒŒμΌμ„ 읽을 수 μ—†μŠ΅λ‹ˆλ‹€ ({', '.join(encodings)})")
144
+ else:
145
+ encodings = ['utf-8', 'cp949', 'euc-kr', 'latin1']
146
+ for encoding in encodings:
147
+ try:
148
+ with open(file.name, 'r', encoding=encoding) as f:
149
+ content = f.read()
150
+ return content, "text"
151
+ except UnicodeDecodeError:
152
+ continue
153
+ raise UnicodeDecodeError(f"❌ μ§€μ›λ˜λŠ” μΈμ½”λ”©μœΌλ‘œ νŒŒμΌμ„ 읽을 수 μ—†μŠ΅λ‹ˆλ‹€ ({', '.join(encodings)})")
154
+ except Exception as e:
155
+ return f"❌ 파일 읽기 였λ₯˜: {str(e)}", "error"
156
+
157
+ def chat(message, history, uploaded_file, system_message="", max_tokens=4000, temperature=0.7, top_p=0.9):
158
+ if not message:
159
+ return "", history
160
+
161
+ system_prefix = """μ €λŠ” μ—¬λŸ¬λΆ„μ˜ μΉœκ·Όν•˜κ³  지적인 AI μ–΄μ‹œμŠ€ν„΄νŠΈ 'GiniGEN'μž…λ‹ˆλ‹€.. λ‹€μŒκ³Ό 같은 μ›μΉ™μœΌλ‘œ μ†Œν†΅ν•˜κ² μŠ΅λ‹ˆλ‹€:
162
+ 1. 🀝 μΉœκ·Όν•˜κ³  곡감적인 νƒœλ„λ‘œ λŒ€ν™”
163
+ 2. πŸ’‘ λͺ…ν™•ν•˜κ³  μ΄ν•΄ν•˜κΈ° μ‰¬μš΄ μ„€λͺ… 제곡
164
+ 3. 🎯 질문의 μ˜λ„λ₯Ό μ •ν™•νžˆ νŒŒμ•…ν•˜μ—¬ λ§žμΆ€ν˜• λ‹΅λ³€
165
+ 4. πŸ“š ν•„μš”ν•œ 경우 μ—…λ‘œλ“œλœ 파일 λ‚΄μš©μ„ μ°Έκ³ ν•˜μ—¬ ꡬ체적인 도움 제곡
166
+ 5. ✨ 좔가적인 톡찰과 μ œμ•ˆμ„ ν†΅ν•œ κ°€μΉ˜ μžˆλŠ” λŒ€ν™”
167
+
168
+ 항상 예의 λ°”λ₯΄κ³  μΉœμ ˆν•˜κ²Œ μ‘λ‹΅ν•˜λ©°, ν•„μš”ν•œ 경우 ꡬ체적인 μ˜ˆμ‹œλ‚˜ μ„€λͺ…을 μΆ”κ°€ν•˜μ—¬
169
+ 이해λ₯Ό λ•κ² μŠ΅λ‹ˆλ‹€."""
170
+
171
+ try:
172
+ # 파일 μ—…λ‘œλ“œ 처리
173
+ if uploaded_file:
174
+ content, file_type = read_uploaded_file(uploaded_file)
175
+ if file_type == "error":
176
+ error_message = content
177
+ chat_history.add_conversation(message, error_message)
178
+ return "", history + [[message, error_message]]
179
+
180
+ file_summary = analyze_file_content(content, file_type)
181
+
182
+ if file_type in ['parquet', 'csv']:
183
+ system_message += f"\n\n파일 λ‚΄μš©:\n```markdown\n{content}\n```"
184
+ else:
185
+ system_message += f"\n\n파일 λ‚΄μš©:\n```\n{content}\n```"
186
+
187
+ if message == "파일 뢄석을 μ‹œμž‘ν•©λ‹ˆλ‹€...":
188
+ message = f"""[파일 ꡬ쑰 뢄석] {file_summary}
189
+ λ‹€μŒ κ΄€μ μ—μ„œ 도움을 λ“œλ¦¬κ² μŠ΅λ‹ˆλ‹€:
190
+ 1. πŸ“‹ μ „λ°˜μ μΈ λ‚΄μš© νŒŒμ•…
191
+ 2. πŸ’‘ μ£Όμš” νŠΉμ§• μ„€λͺ…
192
+ 3. 🎯 μ‹€μš©μ μΈ ν™œμš© λ°©μ•ˆ
193
+ 4. ✨ κ°œμ„  μ œμ•ˆ
194
+ 5. πŸ’¬ μΆ”κ°€ μ§ˆλ¬Έμ΄λ‚˜ ν•„μš”ν•œ μ„€λͺ…"""
195
+
196
+ # λ©”μ‹œμ§€ 처리
197
+ messages = [{"role": "system", "content": system_prefix + system_message}]
198
+
199
+ # 이전 λŒ€ν™” νžˆμŠ€ν† λ¦¬ μΆ”κ°€
200
+ if history:
201
+ for user_msg, assistant_msg in history:
202
+ messages.append({"role": "user", "content": user_msg})
203
+ messages.append({"role": "assistant", "content": assistant_msg})
204
+
205
+ messages.append({"role": "user", "content": message})
206
+
207
+ # API 호좜 및 응닡 처리
208
+ client = get_client()
209
+ partial_message = ""
210
+
211
+ for msg in client.chat_completion(
212
+ messages,
213
+ max_tokens=max_tokens,
214
+ stream=True,
215
+ temperature=temperature,
216
+ top_p=top_p,
217
+ ):
218
+ token = msg.choices[0].delta.get('content', None)
219
+ if token:
220
+ partial_message += token
221
+ current_history = history + [[message, partial_message]]
222
+ yield "", current_history
223
+
224
+ # μ™„μ„±λœ λŒ€ν™” μ €μž₯
225
+ chat_history.add_conversation(message, partial_message)
226
+
227
+ except Exception as e:
228
+ error_msg = f"❌ 였λ₯˜κ°€ λ°œμƒν–ˆμŠ΅λ‹ˆλ‹€: {str(e)}"
229
+ chat_history.add_conversation(message, error_msg)
230
+ yield "", history + [[message, error_msg]]
231
+
232
+ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", title="GiniGEN πŸ€–") as demo:
233
+ # κΈ°μ‘΄ νžˆμŠ€ν† λ¦¬ λ‘œλ“œ
234
+ initial_history = chat_history.format_for_display()
235
+ with gr.Row():
236
+ with gr.Column(scale=2):
237
+ chatbot = gr.Chatbot(
238
+ value=initial_history, # μ €μž₯된 νžˆμŠ€ν† λ¦¬λ‘œ μ΄ˆκΈ°ν™”
239
+ height=600,
240
+ label="λŒ€ν™”μ°½ πŸ’¬",
241
+ show_label=True
242
+ )
243
+
244
+
245
+ msg = gr.Textbox(
246
+ label="λ©”μ‹œμ§€ μž…λ ₯",
247
+ show_label=False,
248
+ placeholder="무엇이든 λ¬Όμ–΄λ³΄μ„Έμš”... πŸ’­",
249
+ container=False
250
+ )
251
+ with gr.Row():
252
+ clear = gr.ClearButton([msg, chatbot], value="λŒ€ν™”λ‚΄μš© μ§€μš°κΈ°")
253
+ send = gr.Button("보내기 πŸ“€")
254
+
255
+ with gr.Column(scale=1):
256
+ gr.Markdown("### GiniGEN πŸ€– [파일 μ—…λ‘œλ“œ] πŸ“\n지원 ν˜•μ‹: ν…μŠ€νŠΈ, μ½”λ“œ, CSV, Parquet 파일")
257
+ file_upload = gr.File(
258
+ label="파일 선택",
259
+ file_types=["text", ".csv", ".parquet"],
260
+ type="filepath"
261
+ )
262
+
263
+ with gr.Accordion("κ³ κΈ‰ μ„€μ • βš™οΈ", open=False):
264
+ system_message = gr.Textbox(label="μ‹œμŠ€ν…œ λ©”μ‹œμ§€ πŸ“", value="")
265
+ max_tokens = gr.Slider(minimum=1, maximum=8000, value=4000, label="μ΅œλŒ€ 토큰 수 πŸ“Š")
266
+ temperature = gr.Slider(minimum=0, maximum=1, value=0.7, label="μ°½μ˜μ„± μˆ˜μ€€ 🌑️")
267
+ top_p = gr.Slider(minimum=0, maximum=1, value=0.9, label="응닡 λ‹€μ–‘μ„± πŸ“ˆ")
268
+
269
+ # μ˜ˆμ‹œ 질문
270
+ gr.Examples(
271
+ examples=[
272
+ ["μ•ˆλ…•ν•˜μ„Έμš”! μ–΄λ–€ 도움이 ν•„μš”ν•˜μ‹ κ°€μš”? 🀝"],
273
+ ["μ œκ°€ μ΄ν•΄ν•˜κΈ° μ‰½κ²Œ μ„€λͺ…ν•΄ μ£Όμ‹œκ² μ–΄μš”? πŸ“š"],
274
+ ["이 λ‚΄μš©μ„ μ‹€μ œλ‘œ μ–΄λ–»κ²Œ ν™œμš©ν•  수 μžˆμ„κΉŒμš”? 🎯"],
275
+ ["μΆ”κ°€λ‘œ μ‘°μ–Έν•΄ μ£Όμ‹€ λ‚΄μš©μ΄ μžˆμœΌμ‹ κ°€μš”? ✨"],
276
+ ["κΆκΈˆν•œ 점이 더 μžˆλŠ”λ° 여쭀봐도 λ κΉŒμš”? πŸ€”"],
277
+ ],
278
+ inputs=msg,
279
+ )
280
+
281
+ # λŒ€ν™”λ‚΄μš© μ§€μš°κΈ° λ²„νŠΌμ— νžˆμŠ€ν† λ¦¬ μ΄ˆκΈ°ν™” κΈ°λŠ₯ μΆ”κ°€
282
+ def clear_chat():
283
+ chat_history.clear_history()
284
+ return None, None
285
+
286
+ # 이벀트 바인딩
287
+ msg.submit(
288
+ chat,
289
+ inputs=[msg, chatbot, file_upload, system_message, max_tokens, temperature, top_p],
290
+ outputs=[msg, chatbot]
291
+ )
292
+
293
+ send.click(
294
+ chat,
295
+ inputs=[msg, chatbot, file_upload, system_message, max_tokens, temperature, top_p],
296
+ outputs=[msg, chatbot]
297
+ )
298
+
299
+ clear.click(
300
+ clear_chat,
301
+ outputs=[msg, chatbot]
302
+ )
303
+
304
+ # 파일 μ—…λ‘œλ“œμ‹œ μžλ™ 뢄석
305
+ file_upload.change(
306
+ lambda: "파일 뢄석을 μ‹œμž‘ν•©λ‹ˆλ‹€...",
307
+ outputs=msg
308
+ ).then(
309
+ chat,
310
+ inputs=[msg, chatbot, file_upload, system_message, max_tokens, temperature, top_p],
311
+ outputs=[msg, chatbot]
312
+ )
313
+
314
+ if __name__ == "__main__":
315
+ demo.launch()