Spaces:
Running
Running
| from transformers import AutoTokenizer, AutoModelForCausalLM | |
| import torch | |
| from datetime import datetime | |
| import os | |
| from mew_log import log_info, log_error # Import your custom logging methods | |
| class HFModel: | |
| def __init__(self, model_name): | |
| parts = model_name.split("/") | |
| self.friendly_name = parts[1] | |
| self.chat_history = [] | |
| self.log_file = f"chat_log_{datetime.now().strftime('%Y%m%d_%H%M%S')}.md" | |
| try: | |
| log_info(f"=== Loading Model: {self.friendly_name} ===") | |
| self.model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, torch_dtype=torch.bfloat16) #.cuda() | |
| self.tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) | |
| log_info(f"=== Model Loaded Successfully: {self.friendly_name} ===") | |
| except Exception as e: | |
| log_error(f"ERROR Loading Model: {e}", e) | |
| raise | |
| def generate_response(self, input_text, max_length=100, skip_special_tokens=True): | |
| try: | |
| inputs = self.tokenizer(input_text, return_tensors="pt").to(self.model.device) | |
| outputs = self.model.generate( | |
| **inputs, | |
| max_length=max_length, | |
| pad_token_id=self.tokenizer.eos_token_id, # Ensure proper padding | |
| do_sample=True, # Enable sampling for more diverse outputs | |
| top_k=50, # Limit sampling to top-k tokens | |
| top_p=0.95, # Use nucleus sampling | |
| temperature=0.7, # Control randomness | |
| ) | |
| #inputs = self.tokenizer(input_text, return_tensors="pt").to(self.model.device) | |
| #outputs = self.model.generate(**inputs, max_length=max_length) | |
| response = self.tokenizer.decode(outputs[0], skip_special_tokens=skip_special_tokens).strip() | |
| log_info(f"Generated Response: {response}") | |
| return response | |
| except Exception as e: | |
| log_error(f"ERROR Generating Response: {e}", e) | |
| raise | |
| def stream_response(self, input_text, max_length=100, skip_special_tokens=True): | |
| try: | |
| inputs = self.tokenizer(input_text, return_tensors="pt").to(self.model.device) | |
| for output in self.model.generate(**inputs, max_length=max_length, do_stream=True): | |
| response = self.tokenizer.decode(output, skip_special_tokens=skip_special_tokens).strip() | |
| yield response | |
| except Exception as e: | |
| log_error(f"ERROR Streaming Response: {e}", e) | |
| raise | |
| def chat(self, user_input, max_length=100, skip_special_tokens=True): | |
| try: | |
| # Add user input to chat history | |
| self.chat_history.append({"role": "user", "content": user_input}) | |
| log_info(f"User Input: {user_input}") | |
| # Generate model response | |
| model_response = self.generate_response(user_input, max_length=max_length, skip_special_tokens=skip_special_tokens) | |
| # Add model response to chat history | |
| self.chat_history.append({"role": "assistant", "content": model_response}) | |
| log_info(f"Assistant Response: {model_response}") | |
| # Save chat log | |
| self.save_chat_log() | |
| return model_response | |
| except Exception as e: | |
| log_error(f"ERROR in Chat: {e}", e) | |
| raise | |
| def save_chat_log(self): | |
| try: | |
| with open(self.log_file, "a", encoding="utf-8") as f: | |
| for entry in self.chat_history[-2:]: # Save only the latest interaction | |
| role = entry["role"] | |
| content = entry["content"] | |
| f.write(f"**{role.capitalize()}:**\n\n{content}\n\n---\n\n") | |
| log_info(f"Chat log saved to {self.log_file}") | |
| except Exception as e: | |
| log_error(f"ERROR Saving Chat Log: {e}", e) | |
| raise | |
| def clear_chat_history(self): | |
| try: | |
| self.chat_history = [] | |
| log_info("Chat history cleared.") | |
| except Exception as e: | |
| log_error(f"ERROR Clearing Chat History: {e}", e) | |
| raise | |
| def print_chat_history(self): | |
| try: | |
| for entry in self.chat_history: | |
| role = entry["role"] | |
| content = entry["content"] | |
| print(f"{role.capitalize()}: {content}\n") | |
| log_info("Printed chat history to console.") | |
| except Exception as e: | |
| log_error(f"ERROR Printing Chat History: {e}", e) | |
| raise |