|
import gradio as gr |
|
from huggingface_hub import InferenceClient |
|
import os |
|
import pandas as pd |
|
from typing import List, Tuple |
|
|
|
|
|
LLM_MODELS = { |
|
"Cohere c4ai-crp-08-2024": "CohereForAI/c4ai-command-r-plus-08-2024", |
|
"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): |
|
"""νμΌ λ΄μ©μ λΆμνμ¬ κ΅¬μ‘°μ μμ½μ λ°ν""" |
|
if file_type in ['parquet', 'csv']: |
|
try: |
|
|
|
lines = content.split('\n') |
|
header = lines[0] |
|
columns = header.count('|') - 1 |
|
rows = len(lines) - 3 |
|
return f"λ°μ΄ν°μ
ꡬ쑰: {columns}κ° μ»¬λΌ, {rows}κ° λ°μ΄ν° μν" |
|
except: |
|
return "λ°μ΄ν°μ
ꡬ쑰 λΆμ μ€ν¨" |
|
|
|
|
|
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"μ½λ ꡬ쑰 λΆμ: μ΄ {total_lines}μ€ (ν¨μ {functions}κ°, ν΄λμ€ {classes}κ°, μν¬νΈ {imports}κ°)" |
|
|
|
paragraphs = content.count('\n\n') + 1 |
|
words = len(content.split()) |
|
return f"λ¬Έμ ꡬ쑰 λΆμ: μ΄ {total_lines}μ€, {paragraphs}κ° λ¬Έλ¨, μ½ {words}κ° λ¨μ΄" |
|
|
|
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': |
|
|
|
encodings = ['utf-8', 'cp949', 'euc-kr', 'latin1'] |
|
for encoding in encodings: |
|
try: |
|
df = pd.read_csv(file.name, encoding=encoding) |
|
content = f"λ°μ΄ν° 미리보기:\n{df.head(10).to_markdown(index=False)}\n\n" |
|
content += f"\nλ°μ΄ν° μ 보:\n" |
|
content += f"- μ΄ ν μ: {len(df)}\n" |
|
content += f"- μ΄ μ΄ μ: {len(df.columns)}\n" |
|
content += f"- μ»¬λΌ λͺ©λ‘: {', '.join(df.columns)}\n" |
|
|
|
content += f"\n컬λΌλ³ λ°μ΄ν° νμ
:\n" |
|
for col, dtype in df.dtypes.items(): |
|
content += f"- {col}: {dtype}\n" |
|
|
|
null_counts = df.isnull().sum() |
|
if null_counts.any(): |
|
content += f"\nκ²°μΈ‘μΉ μ 보:\n" |
|
for col, null_count in null_counts[null_counts > 0].items(): |
|
content += f"- {col}: {null_count}κ°\n" |
|
return content, "csv" |
|
except UnicodeDecodeError: |
|
continue |
|
raise UnicodeDecodeError(f"μ§μλλ μΈμ½λ©({', '.join(encodings)})μΌλ‘ νμΌμ μ½μ μ μμ΅λλ€.") |
|
else: |
|
|
|
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"μ§μλλ μΈμ½λ©({', '.join(encodings)})μΌλ‘ νμΌμ μ½μ μ μμ΅λλ€.") |
|
except Exception as e: |
|
return f"νμΌμ μ½λ μ€ μ€λ₯κ° λ°μνμ΅λλ€: {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 = """λλ νμΌ λΆμ μ λ¬Έκ°μ
λλ€. μ
λ‘λλ νμΌμ λ΄μ©μ κΉμ΄ μκ² λΆμνμ¬ λ€μκ³Ό κ°μ κ΄μ μμ μ€λͺ
ν΄μΌ ν©λλ€: |
|
|
|
1. νμΌμ μ λ°μ μΈ κ΅¬μ‘°μ κ΅¬μ± |
|
2. μ£Όμ λ΄μ©κ³Ό ν¨ν΄ λΆμ |
|
3. λ°μ΄ν°μ νΉμ§κ³Ό μλ―Έ |
|
- λ°μ΄ν°μ
μ κ²½μ°: 컬λΌμ μλ―Έ, λ°μ΄ν° νμ
, κ°μ λΆν¬ |
|
- ν
μ€νΈ/μ½λμ κ²½μ°: ꡬ쑰μ νΉμ§, μ£Όμ ν¨ν΄ |
|
4. μ μ¬μ νμ© λ°©μ |
|
5. λ°μ΄ν° νμ§ λ° κ°μ κ°λ₯ν λΆλΆ |
|
|
|
μ λ¬Έκ°μ κ΄μ μμ μμΈνκ³ κ΅¬μ‘°μ μΈ λΆμμ μ 곡νλ, μ΄ν΄νκΈ° μ½κ² μ€λͺ
νμΈμ. λΆμ κ²°κ³Όλ Markdown νμμΌλ‘ μμ±νκ³ , κ°λ₯ν ν ꡬ체μ μΈ μμλ₯Ό ν¬ν¨νμΈμ.""" |
|
|
|
if uploaded_file: |
|
content, file_type = read_uploaded_file(uploaded_file) |
|
if file_type == "error": |
|
yield "", history + [[message, content]] |
|
return |
|
|
|
|
|
file_summary = analyze_file_content(content, file_type) |
|
|
|
if file_type in ['parquet', 'csv']: |
|
system_message += f"\n\nνμΌ λ΄μ©:\n```markdown\n{content}\n```" |
|
else: |
|
system_message += f"\n\nνμΌ λ΄μ©:\n```\n{content}\n```" |
|
|
|
if message == "νμΌ λΆμμ μμν©λλ€.": |
|
message = f"""[ꡬ쑰 λΆμ] {file_summary} |
|
|
|
λ€μ κ΄μ μμ μμΈ λΆμμ μ 곡ν΄μ£ΌμΈμ: |
|
1. νμΌμ μ λ°μ μΈ κ΅¬μ‘°μ νμ |
|
2. μ£Όμ λ΄μ© λ° κ΅¬μ±μμ λΆμ |
|
3. λ°μ΄ν°/λ΄μ©μ νΉμ§κ³Ό ν¨ν΄ |
|
4. νμ§ λ° μμ±λ νκ° |
|
5. κ°μ κ°λ₯ν λΆλΆ μ μ |
|
6. μ€μ νμ© λ°©μ λ° μΆμ²μ¬ν""" |
|
|
|
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"μΆλ‘ μ€ μ€λ₯κ° λ°μνμ΅λλ€: {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="λ©μμ§λ₯Ό μ
λ ₯νμΈμ", |
|
show_label=False, |
|
placeholder="λ©μμ§λ₯Ό μ
λ ₯νμΈμ...", |
|
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", |
|
label="μ΅μ LLM λͺ¨λΈ μ ν", |
|
info="μ¬μ©ν LLM λͺ¨λΈμ μ ννμΈμ" |
|
) |
|
|
|
file_upload = gr.File( |
|
label="νμΌ μ
λ‘λ (ν
μ€νΈ, μ½λ, CSV, Parquet νμΌ)", |
|
file_types=["text", ".csv", ".parquet"], |
|
type="filepath" |
|
) |
|
|
|
with gr.Accordion("κ³ κΈ μ€μ ", 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") |
|
|
|
|
|
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] |
|
) |
|
|
|
|
|
file_upload.change( |
|
chat, |
|
inputs=[gr.Textbox(value="νμΌ λΆμμ μμν©λλ€."), chatbot, file_upload, model_name, system_message, max_tokens, temperature, top_p], |
|
outputs=[msg, chatbot], |
|
queue=True |
|
) |
|
|
|
|
|
gr.Examples( |
|
examples=[ |
|
["νμΌμ μ λ°μ μΈ κ΅¬μ‘°μ νΉμ§μ μμΈν μ€λͺ
ν΄μ£ΌμΈμ."], |
|
["μ΄ νμΌμ μ£Όμ ν¨ν΄κ³Ό νΉμ§μ λΆμν΄μ£ΌμΈμ."], |
|
["νμΌμ νμ§κ³Ό κ°μ κ°λ₯ν λΆλΆμ νκ°ν΄μ£ΌμΈμ."], |
|
["μ΄ νμΌμ μ€μ λ‘ μ΄λ»κ² νμ©ν μ μμκΉμ?"], |
|
["νμΌμ μ£Όμ λ΄μ©μ μμ½νκ³ ν΅μ¬ μΈμ¬μ΄νΈλ₯Ό λμΆν΄μ£ΌμΈμ."], |
|
["μ΄μ λΆμμ μ΄μ΄μ λ μμΈν μ€λͺ
ν΄μ£ΌμΈμ."], |
|
], |
|
inputs=msg, |
|
) |
|
|
|
if __name__ == "__main__": |
|
demo.launch() |