File size: 10,083 Bytes
f9b9d56 83ee74c 574f73e 705c5b5 83ee74c f9b9d56 ad9db85 79e0b8a ad9db85 f9b9d56 2af89cf da20c1b 99d94e0 da20c1b 99d94e0 da20c1b 2af89cf da20c1b 2af89cf da20c1b 2af89cf da20c1b 2af89cf 705c5b5 0997082 2af89cf 705c5b5 99d94e0 705c5b5 2af89cf 99d94e0 521288b 705c5b5 521288b 705c5b5 2af89cf 0997082 d57197f ad9db85 da20c1b 99d94e0 da20c1b 99d94e0 da20c1b 705c5b5 7ffca43 2af89cf da20c1b 2af89cf da20c1b 2af89cf 7ffca43 99d94e0 7ffca43 da20c1b 7ffca43 da20c1b 2af89cf da20c1b 705c5b5 d57197f 0997082 705c5b5 0997082 ad9db85 da20c1b ad9db85 f2c0975 83ee74c 0997082 83ee74c 7ffca43 705c5b5 da20c1b 0997082 7ffca43 da20c1b 705c5b5 7b3fa19 7ffca43 7b3fa19 63c5e29 ad9db85 7ffca43 63c5e29 2af89cf ad9db85 2011c1c 79e0b8a ad9db85 7ffca43 99d94e0 d57197f 7ffca43 63c5e29 7ffca43 ad9db85 da20c1b 7ffca43 ad9db85 da20c1b 7ffca43 63c5e29 da20c1b 63c5e29 f9b9d56 63c5e29 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 |
import gradio as gr
from huggingface_hub import InferenceClient
import os
import pandas as pd
from typing import List, Tuple
# LLM ๋ชจ๋ธ ์ ์
LLM_MODELS = {
"Cohere c4ai-crp-08-2024": "CohereForAI/c4ai-command-r-plus-08-2024", # ๊ธฐ๋ณธ ๋ชจ๋ธ
"Meta Llama3.3-70B": "meta-llama/Llama-3.3-70B-Instruct",
"Mistral Nemo 2407": "mistralai/Mistral-Nemo-Instruct-2407",
"Alibaba Qwen QwQ-32B": "Qwen/QwQ-32B-Preview"
}
def get_client(model_name):
return InferenceClient(LLM_MODELS[model_name], token=os.getenv("HF_TOKEN"))
def analyze_file_content(content, file_type):
"""ํ์ผ ๋ด์ฉ์ ๋ถ์ํ์ฌ ๊ตฌ์กฐ์ ์์ฝ์ ๋ฐํ"""
if file_type in ['parquet', 'csv']:
try:
# ๋ฐ์ดํฐ์
๊ตฌ์กฐ ๋ถ์
lines = content.split('\n')
header = lines[0]
columns = header.count('|') - 1
rows = len(lines) - 3 # ํค๋์ ๊ตฌ๋ถ์ ์ ์ธ
return f"๋ฐ์ดํฐ์
๊ตฌ์กฐ: {columns}๊ฐ ์ปฌ๋ผ, {rows}๊ฐ ๋ฐ์ดํฐ ์ํ"
except:
return "๋ฐ์ดํฐ์
๊ตฌ์กฐ ๋ถ์ ์คํจ"
# ํ
์คํธ/์ฝ๋ ํ์ผ์ ๊ฒฝ์ฐ
lines = content.split('\n')
total_lines = len(lines)
non_empty_lines = len([line for line in lines if line.strip()])
if any(keyword in content.lower() for keyword in ['def ', 'class ', 'import ', 'function']):
functions = len([line for line in lines if 'def ' in line])
classes = len([line for line in lines if 'class ' in line])
imports = len([line for line in lines if 'import ' in line or 'from ' in line])
return f"์ฝ๋ ๊ตฌ์กฐ ๋ถ์: ์ด {total_lines}์ค (ํจ์ {functions}๊ฐ, ํด๋์ค {classes}๊ฐ, ์ํฌํธ {imports}๊ฐ)"
paragraphs = content.count('\n\n') + 1
words = len(content.split())
return f"๋ฌธ์ ๊ตฌ์กฐ ๋ถ์: ์ด {total_lines}์ค, {paragraphs}๊ฐ ๋ฌธ๋จ, ์ฝ {words}๊ฐ ๋จ์ด"
def read_uploaded_file(file):
if file is None:
return "", ""
try:
file_ext = os.path.splitext(file.name)[1].lower()
if file_ext == '.parquet':
df = pd.read_parquet(file.name, engine='pyarrow')
content = df.head(10).to_markdown(index=False)
return content, "parquet"
elif file_ext == '.csv':
# CSV ํ์ผ ์ฝ๊ธฐ ์ ๋ค์ํ ์ธ์ฝ๋ฉ ์๋
encodings = ['utf-8', 'cp949', 'euc-kr', 'latin1']
for encoding in encodings:
try:
df = pd.read_csv(file.name, encoding=encoding)
content = f"๋ฐ์ดํฐ ๋ฏธ๋ฆฌ๋ณด๊ธฐ:\n{df.head(10).to_markdown(index=False)}\n\n"
content += f"\n๋ฐ์ดํฐ ์ ๋ณด:\n"
content += f"- ์ด ํ ์: {len(df)}\n"
content += f"- ์ด ์ด ์: {len(df.columns)}\n"
content += f"- ์ปฌ๋ผ ๋ชฉ๋ก: {', '.join(df.columns)}\n"
# ๋ฐ์ดํฐ ํ์
์ ๋ณด ์ถ๊ฐ
content += f"\n์ปฌ๋ผ๋ณ ๋ฐ์ดํฐ ํ์
:\n"
for col, dtype in df.dtypes.items():
content += f"- {col}: {dtype}\n"
# ๊ฒฐ์ธก์น ์ ๋ณด ์ถ๊ฐ
null_counts = df.isnull().sum()
if null_counts.any():
content += f"\n๊ฒฐ์ธก์น ์ ๋ณด:\n"
for col, null_count in null_counts[null_counts > 0].items():
content += f"- {col}: {null_count}๊ฐ\n"
return content, "csv"
except UnicodeDecodeError:
continue
raise UnicodeDecodeError(f"์ง์๋๋ ์ธ์ฝ๋ฉ({', '.join(encodings)})์ผ๋ก ํ์ผ์ ์ฝ์ ์ ์์ต๋๋ค.")
else:
# ํ
์คํธ ํ์ผ ์ฝ๊ธฐ ์๋
encodings = ['utf-8', 'cp949', 'euc-kr', 'latin1']
for encoding in encodings:
try:
with open(file.name, 'r', encoding=encoding) as f:
content = f.read()
return content, "text"
except UnicodeDecodeError:
continue
raise UnicodeDecodeError(f"์ง์๋๋ ์ธ์ฝ๋ฉ({', '.join(encodings)})์ผ๋ก ํ์ผ์ ์ฝ์ ์ ์์ต๋๋ค.")
except Exception as e:
return f"ํ์ผ์ ์ฝ๋ ์ค ์ค๋ฅ๊ฐ ๋ฐ์ํ์ต๋๋ค: {str(e)}", "error"
def format_history(history):
formatted_history = []
for user_msg, assistant_msg in history:
formatted_history.append({"role": "user", "content": user_msg})
if assistant_msg:
formatted_history.append({"role": "assistant", "content": assistant_msg})
return formatted_history
def chat(message, history, uploaded_file, model_name, system_message="", max_tokens=4000, temperature=0.7, top_p=0.9):
system_prefix = """๋๋ ํ์ผ ๋ถ์ ์ ๋ฌธ๊ฐ์
๋๋ค. ์
๋ก๋๋ ํ์ผ์ ๋ด์ฉ์ ๊น์ด ์๊ฒ ๋ถ์ํ์ฌ ๋ค์๊ณผ ๊ฐ์ ๊ด์ ์์ ์ค๋ช
ํด์ผ ํฉ๋๋ค:
1. ํ์ผ์ ์ ๋ฐ์ ์ธ ๊ตฌ์กฐ์ ๊ตฌ์ฑ
2. ์ฃผ์ ๋ด์ฉ๊ณผ ํจํด ๋ถ์
3. ๋ฐ์ดํฐ์ ํน์ง๊ณผ ์๋ฏธ
- ๋ฐ์ดํฐ์
์ ๊ฒฝ์ฐ: ์ปฌ๋ผ์ ์๋ฏธ, ๋ฐ์ดํฐ ํ์
, ๊ฐ์ ๋ถํฌ
- ํ
์คํธ/์ฝ๋์ ๊ฒฝ์ฐ: ๊ตฌ์กฐ์ ํน์ง, ์ฃผ์ ํจํด
4. ์ ์ฌ์ ํ์ฉ ๋ฐฉ์
5. ๋ฐ์ดํฐ ํ์ง ๋ฐ ๊ฐ์ ๊ฐ๋ฅํ ๋ถ๋ถ
์ ๋ฌธ๊ฐ์ ๊ด์ ์์ ์์ธํ๊ณ ๊ตฌ์กฐ์ ์ธ ๋ถ์์ ์ ๊ณตํ๋, ์ดํดํ๊ธฐ ์ฝ๊ฒ ์ค๋ช
ํ์ธ์. ๋ถ์ ๊ฒฐ๊ณผ๋ Markdown ํ์์ผ๋ก ์์ฑํ๊ณ , ๊ฐ๋ฅํ ํ ๊ตฌ์ฒด์ ์ธ ์์๋ฅผ ํฌํจํ์ธ์."""
if uploaded_file:
content, file_type = read_uploaded_file(uploaded_file)
if file_type == "error":
yield "", history + [[message, content]]
return
# ํ์ผ ๋ด์ฉ ๋ถ์ ๋ฐ ๊ตฌ์กฐ์ ์์ฝ
file_summary = analyze_file_content(content, file_type)
if file_type in ['parquet', 'csv']:
system_message += f"\n\nํ์ผ ๋ด์ฉ:\n```markdown\n{content}\n```"
else:
system_message += f"\n\nํ์ผ ๋ด์ฉ:\n```\n{content}\n```"
if message == "ํ์ผ ๋ถ์์ ์์ํฉ๋๋ค.":
message = f"""[๊ตฌ์กฐ ๋ถ์] {file_summary}
๋ค์ ๊ด์ ์์ ์์ธ ๋ถ์์ ์ ๊ณตํด์ฃผ์ธ์:
1. ํ์ผ์ ์ ๋ฐ์ ์ธ ๊ตฌ์กฐ์ ํ์
2. ์ฃผ์ ๋ด์ฉ ๋ฐ ๊ตฌ์ฑ์์ ๋ถ์
3. ๋ฐ์ดํฐ/๋ด์ฉ์ ํน์ง๊ณผ ํจํด
4. ํ์ง ๋ฐ ์์ฑ๋ ํ๊ฐ
5. ๊ฐ์ ๊ฐ๋ฅํ ๋ถ๋ถ ์ ์
6. ์ค์ ํ์ฉ ๋ฐฉ์ ๋ฐ ์ถ์ฒ์ฌํญ"""
messages = [{"role": "system", "content": f"{system_prefix} {system_message}"}]
messages.extend(format_history(history))
messages.append({"role": "user", "content": message})
try:
client = get_client(model_name)
partial_message = ""
for msg in client.chat_completion(
messages,
max_tokens=max_tokens,
stream=True,
temperature=temperature,
top_p=top_p,
):
token = msg.choices[0].delta.get('content', None)
if token:
partial_message += token
yield "", history + [[message, partial_message]]
except Exception as e:
error_msg = f"์ถ๋ก ์ค ์ค๋ฅ๊ฐ ๋ฐ์ํ์ต๋๋ค: {str(e)}"
yield "", history + [[message, error_msg]]
css = """
footer {visibility: hidden}
"""
with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
with gr.Row():
with gr.Column(scale=2):
chatbot = gr.Chatbot(height=600)
msg = gr.Textbox(
label="๋ฉ์์ง๋ฅผ ์
๋ ฅํ์ธ์",
show_label=False,
placeholder="๋ฉ์์ง๋ฅผ ์
๋ ฅํ์ธ์...",
container=False
)
clear = gr.ClearButton([msg, chatbot])
with gr.Column(scale=1):
model_name = gr.Radio(
choices=list(LLM_MODELS.keys()),
value="Cohere c4ai-crp-08-2024", # ๊ธฐ๋ณธ๊ฐ์ Cohere ๋ชจ๋ธ๋ก ๋ช
์์ ์ง์
label="์ต์ LLM ๋ชจ๋ธ ์ ํ",
info="์ฌ์ฉํ LLM ๋ชจ๋ธ์ ์ ํํ์ธ์"
)
file_upload = gr.File(
label="ํ์ผ ์
๋ก๋ (ํ
์คํธ, ์ฝ๋, CSV, Parquet ํ์ผ)",
file_types=["text", ".csv", ".parquet"],
type="filepath"
)
with gr.Accordion("๊ณ ๊ธ ์ค์ ", open=False):
system_message = gr.Textbox(label="System Message", value="")
max_tokens = gr.Slider(minimum=1, maximum=8000, value=4000, label="Max Tokens")
temperature = gr.Slider(minimum=0, maximum=1, value=0.7, label="Temperature")
top_p = gr.Slider(minimum=0, maximum=1, value=0.9, label="Top P")
# ์ด๋ฒคํธ ๋ฐ์ธ๋ฉ
msg.submit(
chat,
inputs=[msg, chatbot, file_upload, model_name, system_message, max_tokens, temperature, top_p],
outputs=[msg, chatbot],
queue=True
).then(
lambda: gr.update(interactive=True),
None,
[msg]
)
# ํ์ผ ์
๋ก๋ ์ ์๋ ๋ถ์
file_upload.change(
chat,
inputs=[gr.Textbox(value="ํ์ผ ๋ถ์์ ์์ํฉ๋๋ค."), chatbot, file_upload, model_name, system_message, max_tokens, temperature, top_p],
outputs=[msg, chatbot],
queue=True
)
# ์์ ์ถ๊ฐ
gr.Examples(
examples=[
["ํ์ผ์ ์ ๋ฐ์ ์ธ ๊ตฌ์กฐ์ ํน์ง์ ์์ธํ ์ค๋ช
ํด์ฃผ์ธ์."],
["์ด ํ์ผ์ ์ฃผ์ ํจํด๊ณผ ํน์ง์ ๋ถ์ํด์ฃผ์ธ์."],
["ํ์ผ์ ํ์ง๊ณผ ๊ฐ์ ๊ฐ๋ฅํ ๋ถ๋ถ์ ํ๊ฐํด์ฃผ์ธ์."],
["์ด ํ์ผ์ ์ค์ ๋ก ์ด๋ป๊ฒ ํ์ฉํ ์ ์์๊น์?"],
["ํ์ผ์ ์ฃผ์ ๋ด์ฉ์ ์์ฝํ๊ณ ํต์ฌ ์ธ์ฌ์ดํธ๋ฅผ ๋์ถํด์ฃผ์ธ์."],
["์ด์ ๋ถ์์ ์ด์ด์ ๋ ์์ธํ ์ค๋ช
ํด์ฃผ์ธ์."],
],
inputs=msg,
)
if __name__ == "__main__":
demo.launch() |