# ---------------------------------------------------------------
# Host path to the quantized model directory
# ---------------------------------------------------------------
MODEL_HOST_PATH=/path/to/Qwen3.5-122B-A10B-NVFP4

# ---------------------------------------------------------------
# API port exposed on the host
# ---------------------------------------------------------------
HOST_PORT=8000

# ---------------------------------------------------------------
# vLLM serving parameters  (tune to your hardware)
# ---------------------------------------------------------------
MAX_MODEL_LEN=262144
MAX_NUM_SEQS=4
GPU_MEMORY_UTILIZATION=0.9
SWAP_SPACE=0
MAX_NUM_BATCHED_TOKENS=32768