# vLLM model config.
MODEL_NAME=${MODEL_NAME:-meta-llama/Llama-3.1-8B-Instruct}
MAX_MODEL_LEN=${MAX_MODEL_LEN:-8192}
# QUANTIZATION: none | awq | gptq | fp8  (support differs by GPU vendor;
# see docs/retro/compatibility-matrix.md)
QUANTIZATION=${QUANTIZATION:-none}