StoryStar / app.py
openfree's picture
Update app.py
2ff143a verified
raw
history blame
9.99 kB
import gradio as gr
from huggingface_hub import InferenceClient
import os
import pandas as pd
from typing import List, Tuple
# LLM Models Definition
LLM_MODELS = {
"Cohere c4ai-crp-08-2024": "CohereForAI/c4ai-command-r-plus-08-2024", # Default
"Meta Llama3.3-70B": "meta-llama/Llama-3.3-70B-Instruct" # Backup model
}
def get_client(model_name="Cohere c4ai-crp-08-2024"):
try:
return InferenceClient(LLM_MODELS[model_name], token=os.getenv("HF_TOKEN"))
except Exception:
# If primary model fails, try backup model
return InferenceClient(LLM_MODELS["Meta Llama3.3-70B"], token=os.getenv("HF_TOKEN"))
def analyze_file_content(content, file_type):
"""Analyze file content and return structural summary"""
if file_type in ['parquet', 'csv']:
try:
lines = content.split('\n')
header = lines[0]
columns = header.count('|') - 1
rows = len(lines) - 3
return f"πŸ“Š Dataset Structure: {columns} columns, {rows} data samples"
except:
return "❌ Dataset structure analysis failed"
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"πŸ’» Code Structure: {total_lines} lines (Functions: {functions}, Classes: {classes}, Imports: {imports})"
paragraphs = content.count('\n\n') + 1
words = len(content.split())
return f"πŸ“ Document Structure: {total_lines} lines, {paragraphs} paragraphs, ~{words} words"
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"πŸ“Š Data Preview:\n{df.head(10).to_markdown(index=False)}\n\n"
content += f"\nπŸ“ˆ Data Information:\n"
content += f"- Total Rows: {len(df)}\n"
content += f"- Total Columns: {len(df.columns)}\n"
content += f"- Column List: {', '.join(df.columns)}\n"
content += f"\nπŸ“‹ Column Data Types:\n"
for col, dtype in df.dtypes.items():
content += f"- {col}: {dtype}\n"
null_counts = df.isnull().sum()
if null_counts.any():
content += f"\n⚠️ Missing Values:\n"
for col, null_count in null_counts[null_counts > 0].items():
content += f"- {col}: {null_count} missing\n"
return content, "csv"
except UnicodeDecodeError:
continue
raise UnicodeDecodeError(f"❌ Unable to read file with supported encodings ({', '.join(encodings)})")
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"❌ Unable to read file with supported encodings ({', '.join(encodings)})")
except Exception as e:
return f"❌ Error reading file: {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
# μ‹œμŠ€ν…œ ν”„λ‘¬ν”„νŠΈ μˆ˜μ •
def chat(message, history, uploaded_file, system_message="", max_tokens=4000, temperature=0.7, top_p=0.9):
system_prefix = """μ €λŠ” μ—¬λŸ¬λΆ„μ˜ μΉœκ·Όν•˜κ³  지적인 AI μ–΄μ‹œμŠ€ν„΄νŠΈμž…λ‹ˆλ‹€. λ‹€μŒκ³Ό 같은 μ›μΉ™μœΌλ‘œ μ†Œν†΅ν•˜κ² μŠ΅λ‹ˆλ‹€:
1. 🀝 μΉœκ·Όν•˜κ³  곡감적인 νƒœλ„λ‘œ λŒ€ν™”
2. πŸ’‘ λͺ…ν™•ν•˜κ³  μ΄ν•΄ν•˜κΈ° μ‰¬μš΄ μ„€λͺ… 제곡
3. 🎯 질문의 μ˜λ„λ₯Ό μ •ν™•νžˆ νŒŒμ•…ν•˜μ—¬ λ§žμΆ€ν˜• λ‹΅λ³€
4. πŸ“š ν•„μš”ν•œ 경우 μ—…λ‘œλ“œλœ 파일 λ‚΄μš©μ„ μ°Έκ³ ν•˜μ—¬ ꡬ체적인 도움 제곡
5. ✨ 좔가적인 톡찰과 μ œμ•ˆμ„ ν†΅ν•œ κ°€μΉ˜ μžˆλŠ” λŒ€ν™”
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—¬
이해λ₯Ό λ•κ² μŠ΅λ‹ˆλ‹€."""
if uploaded_file:
content, file_type = read_uploaded_file(uploaded_file)
if file_type == "error":
return "", [{"role": "user", "content": message}, {"role": "assistant", "content": content}]
file_summary = analyze_file_content(content, file_type)
if file_type in ['parquet', 'csv']:
system_message += f"\n\n파일 λ‚΄μš©:\n```markdown\n{content}\n```"
else:
system_message += f"\n\n파일 λ‚΄μš©:\n```\n{content}\n```"
if message == "Starting file analysis...":
message = f"""[파일 ꡬ쑰 뢄석] {file_summary}
λ‹€μŒ κ΄€μ μ—μ„œ 도움을 λ“œλ¦¬κ² μŠ΅λ‹ˆλ‹€:
1. πŸ“‹ μ „λ°˜μ μΈ λ‚΄μš© νŒŒμ•…
2. πŸ’‘ μ£Όμš” νŠΉμ§• μ„€λͺ…
3. 🎯 μ‹€μš©μ μΈ ν™œμš© λ°©μ•ˆ
4. ✨ κ°œμ„  μ œμ•ˆ
5. πŸ’¬ μΆ”κ°€ μ§ˆλ¬Έμ΄λ‚˜ ν•„μš”ν•œ μ„€λͺ…"""
messages = [{"role": "system", "content": f"{system_prefix} {system_message}"}]
if history is not None:
for item in history:
if isinstance(item, dict):
messages.append(item)
elif isinstance(item, (list, tuple)) and len(item) == 2:
messages.append({"role": "user", "content": item[0]})
if item[1]:
messages.append({"role": "assistant", "content": item[1]})
messages.append({"role": "user", "content": message})
try:
client = get_client()
partial_message = ""
current_history = []
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
current_history = [
{"role": "user", "content": message},
{"role": "assistant", "content": partial_message}
]
yield "", current_history
except Exception as e:
error_msg = f"❌ 였λ₯˜κ°€ λ°œμƒν–ˆμŠ΅λ‹ˆλ‹€: {str(e)}"
error_history = [
{"role": "user", "content": message},
{"role": "assistant", "content": error_msg}
]
yield "", error_history
# UI ν…μŠ€νŠΈ ν•œκΈ€ν™”
with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", title="GiniGEN πŸ€–") as demo:
gr.HTML(
"""
<div style="text-align: center; max-width: 800px; margin: 0 auto;">
<h1 style="font-size: 3em; font-weight: 600; margin: 0.5em;">AI μ–΄μ‹œμŠ€ν„΄νŠΈ πŸ€–</h1>
<h3 style="font-size: 1.2em; margin: 1em;">λ‹Ήμ‹ μ˜ λ“ λ“ ν•œ λŒ€ν™” νŒŒνŠΈλ„ˆ πŸ’¬</h3>
</div>
"""
)
with gr.Row():
with gr.Column(scale=2):
chatbot = gr.Chatbot(
height=600,
label="λŒ€ν™”μ°½ πŸ’¬",
type="messages"
)
msg = gr.Textbox(
label="λ©”μ‹œμ§€ μž…λ ₯",
show_label=False,
placeholder="무엇이든 λ¬Όμ–΄λ³΄μ„Έμš”... πŸ’­",
container=False
)
with gr.Row():
clear = gr.ClearButton([msg, chatbot], value="λŒ€ν™”λ‚΄μš© μ§€μš°κΈ°")
send = gr.Button("보내기 πŸ“€")
with gr.Column(scale=1):
gr.Markdown("### 파일 μ—…λ‘œλ“œ πŸ“\n지원 ν˜•μ‹: ν…μŠ€νŠΈ, μ½”λ“œ, CSV, Parquet 파일")
file_upload = gr.File(
label="파일 선택",
file_types=["text", ".csv", ".parquet"],
type="filepath"
)
with gr.Accordion("κ³ κΈ‰ μ„€μ • βš™οΈ", open=False):
system_message = gr.Textbox(label="μ‹œμŠ€ν…œ λ©”μ‹œμ§€ πŸ“", value="")
max_tokens = gr.Slider(minimum=1, maximum=8000, value=4000, label="μ΅œλŒ€ 토큰 수 πŸ“Š")
temperature = gr.Slider(minimum=0, maximum=1, value=0.7, label="μ°½μ˜μ„± μˆ˜μ€€ 🌑️")
top_p = gr.Slider(minimum=0, maximum=1, value=0.9, label="응닡 λ‹€μ–‘μ„± πŸ“ˆ")
# μ˜ˆμ‹œ 질문 μˆ˜μ •
gr.Examples(
examples=[
["μ•ˆλ…•ν•˜μ„Έμš”! μ–΄λ–€ 도움이 ν•„μš”ν•˜μ‹ κ°€μš”? 🀝"],
["이 λ‚΄μš©μ— λŒ€ν•΄ μ’€ 더 μžμ„Ένžˆ μ„€λͺ…ν•΄ μ£Όμ‹€ 수 μžˆλ‚˜μš”? πŸ’‘"],
["μ œκ°€ μ΄ν•΄ν•˜κΈ° μ‰½κ²Œ μ„€λͺ…ν•΄ μ£Όμ‹œκ² μ–΄μš”? πŸ“š"],
["이 λ‚΄μš©μ„ μ‹€μ œλ‘œ μ–΄λ–»κ²Œ ν™œμš©ν•  수 μžˆμ„κΉŒμš”? 🎯"],
["μΆ”κ°€λ‘œ μ‘°μ–Έν•΄ μ£Όμ‹€ λ‚΄μš©μ΄ μžˆμœΌμ‹ κ°€μš”? ✨"],
["κΆκΈˆν•œ 점이 더 μžˆλŠ”λ° 여쭀봐도 λ κΉŒμš”? πŸ€”"],
],
inputs=msg,
)
if __name__ == "__main__":
demo.launch()