MODEL_ID = google/medgemma-27b-text-it
PYTORCH_CUDA_ALLOC_CONF = expandable_segments:True,max_split_size_mb:64
FORCE_CPU_LLM =
MODEL_FALLBACK_ID = google/medgemma-4b-it
QUANT_MODE = 4bit
HF_HUB_ENABLE_HF_TRANSFER = 1
HF_HOME = /data/econsult/hf_cache
TRANSFORMERS_CACHE = /data/econsult/hf_cache