Spaces:
Running
Running
| # storyverse_weaver/core/llm_services.py | |
| import os | |
| import google.generativeai as genai | |
| from huggingface_hub import InferenceClient | |
| # from dotenv import load_dotenv # Optional: for local .env file | |
| # load_dotenv() # Load environment variables from .env file if present | |
| GOOGLE_API_KEY = os.getenv("STORYVERSE_GOOGLE_API_KEY") # Use specific env var names | |
| HF_TOKEN = os.getenv("STORYVERSE_HF_TOKEN") | |
| GEMINI_TEXT_CONFIGURED = False | |
| HF_TEXT_CONFIGURED = False | |
| hf_inference_text_client = None | |
| class LLMTextResponse: | |
| def __init__(self, text=None, error=None, success=True, model_id_used="unknown_text_llm"): | |
| self.text, self.error, self.success, self.model_id_used = text, error, success, model_id_used | |
| def initialize_text_llms(): | |
| global GOOGLE_API_KEY, HF_TOKEN, GEMINI_TEXT_CONFIGURED, HF_TEXT_CONFIGURED, hf_inference_text_client | |
| print("INFO: llm_services.py - Initializing Text LLM clients...") | |
| if GOOGLE_API_KEY and GOOGLE_API_KEY.strip(): | |
| try: | |
| genai.configure(api_key=GOOGLE_API_KEY) | |
| GEMINI_TEXT_CONFIGURED = True | |
| print("SUCCESS: llm_services.py - Google Gemini API (for text) configured.") | |
| except Exception as e: | |
| print(f"ERROR: llm_services.py - Failed to configure Google Gemini API: {e}") | |
| GEMINI_TEXT_CONFIGURED = False | |
| else: | |
| print("WARNING: llm_services.py - STORYVERSE_GOOGLE_API_KEY not found or empty.") | |
| GEMINI_TEXT_CONFIGURED = False | |
| if HF_TOKEN and HF_TOKEN.strip(): | |
| try: | |
| hf_inference_text_client = InferenceClient(token=HF_TOKEN) | |
| HF_TEXT_CONFIGURED = True | |
| print("SUCCESS: llm_services.py - Hugging Face InferenceClient (for text) initialized.") | |
| except Exception as e: | |
| print(f"ERROR: llm_services.py - Failed to initialize HF InferenceClient: {e}") | |
| HF_TEXT_CONFIGURED = False | |
| else: | |
| print("WARNING: llm_services.py - STORYVERSE_HF_TOKEN not found or empty.") | |
| HF_TEXT_CONFIGURED = False | |
| print(f"INFO: llm_services.py - Text LLM Init complete. Gemini Text: {GEMINI_TEXT_CONFIGURED}, HF Text: {HF_TEXT_CONFIGURED}") | |
| def is_gemini_text_ready(): return GEMINI_TEXT_CONFIGURED | |
| def is_hf_text_ready(): return HF_TEXT_CONFIGURED | |
| def generate_text_gemini(prompt: str, model_id: str = "gemini-1.5-flash-latest", system_prompt: str = None, temperature: float = 0.7, max_tokens: int = 512) -> LLMTextResponse: | |
| if not is_gemini_text_ready(): | |
| return LLMTextResponse(error="Gemini text API not configured.", success=False, model_id_used=model_id) | |
| try: | |
| model = genai.GenerativeModel(model_name=model_id, system_instruction=system_prompt) | |
| config = genai.types.GenerationConfig(temperature=temperature, max_output_tokens=max_tokens) | |
| response = model.generate_content(prompt, generation_config=config) | |
| # Add robust response checking as in AlgoForge's llm_clients.py | |
| if response.prompt_feedback and response.prompt_feedback.block_reason: | |
| return LLMTextResponse(error=f"Gemini: Prompt blocked ({response.prompt_feedback.block_reason})", success=False, model_id_used=model_id) | |
| if not response.candidates or not response.candidates[0].content.parts: | |
| return LLMTextResponse(error=f"Gemini: No content generated (Finish reason: {response.candidates[0].finish_reason if response.candidates else 'Unknown'})", success=False, model_id_used=model_id) | |
| return LLMTextResponse(text=response.text, model_id_used=model_id) | |
| except Exception as e: | |
| return LLMTextResponse(error=f"Gemini API Error ({model_id}): {type(e).__name__} - {str(e)}", success=False, model_id_used=model_id) | |
| def generate_text_hf(prompt: str, model_id: str = "mistralai/Mistral-7B-Instruct-v0.2", system_prompt: str = None, temperature: float = 0.7, max_tokens: int = 512) -> LLMTextResponse: | |
| if not is_hf_text_ready() or not hf_inference_text_client: | |
| return LLMTextResponse(error="HF text API not configured.", success=False, model_id_used=model_id) | |
| full_prompt = f"<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n{prompt} [/INST]" if system_prompt else prompt | |
| try: | |
| use_sample = temperature > 0.001 | |
| response_text = hf_inference_text_client.text_generation( | |
| full_prompt, model=model_id, max_new_tokens=max_tokens, | |
| temperature=temperature if use_sample else None, do_sample=use_sample | |
| ) | |
| return LLMTextResponse(text=response_text, model_id_used=model_id) | |
| except Exception as e: | |
| return LLMTextResponse(error=f"HF API Error ({model_id}): {type(e).__name__} - {str(e)}", success=False, model_id_used=model_id) | |
| print("DEBUG: core.llm_services (for StoryVerseWeaver) - Module defined.")# storyverse_weaver/core/llm_services.py | |
| import os | |
| import google.generativeai as genai | |
| from huggingface_hub import InferenceClient | |
| # from dotenv import load_dotenv # Optional: for local .env file | |
| # load_dotenv() # Load environment variables from .env file if present | |
| GOOGLE_API_KEY = os.getenv("STORYVERSE_GOOGLE_API_KEY") # Use specific env var names | |
| HF_TOKEN = os.getenv("STORYVERSE_HF_TOKEN") | |
| GEMINI_TEXT_CONFIGURED = False | |
| HF_TEXT_CONFIGURED = False | |
| hf_inference_text_client = None | |
| class LLMTextResponse: | |
| def __init__(self, text=None, error=None, success=True, model_id_used="unknown_text_llm"): | |
| self.text, self.error, self.success, self.model_id_used = text, error, success, model_id_used | |
| def initialize_text_llms(): | |
| global GOOGLE_API_KEY, HF_TOKEN, GEMINI_TEXT_CONFIGURED, HF_TEXT_CONFIGURED, hf_inference_text_client | |
| print("INFO: llm_services.py - Initializing Text LLM clients...") | |
| if GOOGLE_API_KEY and GOOGLE_API_KEY.strip(): | |
| try: | |
| genai.configure(api_key=GOOGLE_API_KEY) | |
| GEMINI_TEXT_CONFIGURED = True | |
| print("SUCCESS: llm_services.py - Google Gemini API (for text) configured.") | |
| except Exception as e: | |
| print(f"ERROR: llm_services.py - Failed to configure Google Gemini API: {e}") | |
| GEMINI_TEXT_CONFIGURED = False | |
| else: | |
| print("WARNING: llm_services.py - STORYVERSE_GOOGLE_API_KEY not found or empty.") | |
| GEMINI_TEXT_CONFIGURED = False | |
| if HF_TOKEN and HF_TOKEN.strip(): | |
| try: | |
| hf_inference_text_client = InferenceClient(token=HF_TOKEN) | |
| HF_TEXT_CONFIGURED = True | |
| print("SUCCESS: llm_services.py - Hugging Face InferenceClient (for text) initialized.") | |
| except Exception as e: | |
| print(f"ERROR: llm_services.py - Failed to initialize HF InferenceClient: {e}") | |
| HF_TEXT_CONFIGURED = False | |
| else: | |
| print("WARNING: llm_services.py - STORYVERSE_HF_TOKEN not found or empty.") | |
| HF_TEXT_CONFIGURED = False | |
| print(f"INFO: llm_services.py - Text LLM Init complete. Gemini Text: {GEMINI_TEXT_CONFIGURED}, HF Text: {HF_TEXT_CONFIGURED}") | |
| def is_gemini_text_ready(): return GEMINI_TEXT_CONFIGURED | |
| def is_hf_text_ready(): return HF_TEXT_CONFIGURED | |
| def generate_text_gemini(prompt: str, model_id: str = "gemini-1.5-flash-latest", system_prompt: str = None, temperature: float = 0.7, max_tokens: int = 512) -> LLMTextResponse: | |
| if not is_gemini_text_ready(): | |
| return LLMTextResponse(error="Gemini text API not configured.", success=False, model_id_used=model_id) | |
| try: | |
| model = genai.GenerativeModel(model_name=model_id, system_instruction=system_prompt) | |
| config = genai.types.GenerationConfig(temperature=temperature, max_output_tokens=max_tokens) | |
| response = model.generate_content(prompt, generation_config=config) | |
| # Add robust response checking as in AlgoForge's llm_clients.py | |
| if response.prompt_feedback and response.prompt_feedback.block_reason: | |
| return LLMTextResponse(error=f"Gemini: Prompt blocked ({response.prompt_feedback.block_reason})", success=False, model_id_used=model_id) | |
| if not response.candidates or not response.candidates[0].content.parts: | |
| return LLMTextResponse(error=f"Gemini: No content generated (Finish reason: {response.candidates[0].finish_reason if response.candidates else 'Unknown'})", success=False, model_id_used=model_id) | |
| return LLMTextResponse(text=response.text, model_id_used=model_id) | |
| except Exception as e: | |
| return LLMTextResponse(error=f"Gemini API Error ({model_id}): {type(e).__name__} - {str(e)}", success=False, model_id_used=model_id) | |
| def generate_text_hf(prompt: str, model_id: str = "mistralai/Mistral-7B-Instruct-v0.2", system_prompt: str = None, temperature: float = 0.7, max_tokens: int = 512) -> LLMTextResponse: | |
| if not is_hf_text_ready() or not hf_inference_text_client: | |
| return LLMTextResponse(error="HF text API not configured.", success=False, model_id_used=model_id) | |
| full_prompt = f"<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n{prompt} [/INST]" if system_prompt else prompt | |
| try: | |
| use_sample = temperature > 0.001 | |
| response_text = hf_inference_text_client.text_generation( | |
| full_prompt, model=model_id, max_new_tokens=max_tokens, | |
| temperature=temperature if use_sample else None, do_sample=use_sample | |
| ) | |
| return LLMTextResponse(text=response_text, model_id_used=model_id) | |
| except Exception as e: | |
| return LLMTextResponse(error=f"HF API Error ({model_id}): {type(e).__name__} - {str(e)}", success=False, model_id_used=model_id) | |
| print("DEBUG: core.llm_services (for StoryVerseWeaver) - Module defined.") |