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"π Dataset Structure: {columns} columns, {rows} data samples" except: return "β 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"π» Code Structure: {total_lines} lines (Functions: {functions}, Classes: {classes}, Imports: {imports})" paragraphs = content.count('\n\n') + 1 words = len(content.split()) return f"π 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"π Data Preview:\n{df.head(10).to_markdown(index=False)}\n\n" content += f"\nπ 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"\nπ 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"\nβ οΈ 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"β 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"β Unable to read file with supported encodings ({', '.join(encodings)})") except Exception as e: return f"β 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 # μμ€ν ν둬ννΈ μμ def chat(message, history, uploaded_file, system_message="", max_tokens=4000, temperature=0.7, top_p=0.9): system_prefix = """μ λ μ¬λ¬λΆμ μΉκ·Όνκ³ μ§μ μΈ AI μ΄μμ€ν΄νΈμ λλ€. λ€μκ³Ό κ°μ μμΉμΌλ‘ μν΅νκ² μ΅λλ€: 1. π€ μΉκ·Όνκ³ κ³΅κ°μ μΈ νλλ‘ λν 2. π‘ λͺ ννκ³ μ΄ν΄νκΈ° μ¬μ΄ μ€λͺ μ 곡 3. π― μ§λ¬Έμ μλλ₯Ό μ νν νμ νμ¬ λ§μΆ€ν λ΅λ³ 4. π νμν κ²½μ° μ λ‘λλ νμΌ λ΄μ©μ μ°Έκ³ νμ¬ κ΅¬μ²΄μ μΈ λμ μ 곡 5. β¨ μΆκ°μ μΈ ν΅μ°°κ³Ό μ μμ ν΅ν κ°μΉ μλ λν νμ μμ λ°λ₯΄κ³ μΉμ νκ² μλ΅νλ©°, νμν κ²½μ° κ΅¬μ²΄μ μΈ μμλ μ€λͺ μ μΆκ°νμ¬ μ΄ν΄λ₯Ό λκ² μ΅λλ€.""" 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\nνμΌ λ΄μ©:\n```markdown\n{content}\n```" else: system_message += f"\n\nνμΌ λ΄μ©:\n```\n{content}\n```" if message == "Starting file analysis...": message = f"""[νμΌ κ΅¬μ‘° λΆμ] {file_summary} λ€μ κ΄μ μμ λμμ λλ¦¬κ² μ΅λλ€: 1. π μ λ°μ μΈ λ΄μ© νμ 2. π‘ μ£Όμ νΉμ§ μ€λͺ 3. π― μ€μ©μ μΈ νμ© λ°©μ 4. β¨ κ°μ μ μ 5. π¬ μΆκ° μ§λ¬Έμ΄λ νμν 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"β μ€λ₯κ° λ°μνμ΅λλ€: {str(e)}" error_history = [ {"role": "user", "content": message}, {"role": "assistant", "content": error_msg} ] yield "", error_history # UI ν μ€νΈ νκΈν with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", title="GiniGEN π€") as demo: gr.HTML( """