import os MODEL_REPO_ID = os.getenv( "MODEL_REPO_ID", "mjpsm/activity-progression-model-600-qwen3.5-0.8b-gguf", ) MODEL_FILENAME = os.getenv( "MODEL_FILENAME", "activity-progression-model-600-qwen3.5-0.8b-Q8_0.gguf", ) MODEL_VERSION = os.getenv("MODEL_VERSION", "v1") QUANTIZATION = "Q8_0" RUNTIME_NAME = "llama-cpp-python" PROMPT_FORMAT = "raw_sft_completion" # Optional. When unset, huggingface_hub falls back to the token stored by # `hf auth login`, which is what local development uses. HF_TOKEN = os.getenv("HF_TOKEN") LLAMA_CONTEXT_SIZE = int(os.getenv("LLAMA_CONTEXT_SIZE", "2048")) LLAMA_THREADS = int(os.getenv("LLAMA_THREADS", "2")) # 0 = CPU only. Set to -1 to offload every layer (e.g. Metal on Apple Silicon). LLAMA_GPU_LAYERS = int(os.getenv("LLAMA_GPU_LAYERS", "0")) # Internal generation settings. These match the settings the model was # evaluated with and are deliberately NOT part of the public request schema. MAX_NEW_TOKENS = 450 TEMPERATURE = 0.0 TOP_P = 1.0 REPEAT_PENALTY = 1.08 SEED = 42