Veblen34's picture
Upload folder using huggingface_hub
cd6775d verified
Raw
History Blame Contribute Delete
486 Bytes
#!/bin/bash
# [ragqa] vLLM 服务(仅基座,无 LoRA,端口 50001)。需要 RAGQA_LLAMA_MODEL。
export CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES:-0,1}
: "${RAGQA_LLAMA_MODEL:?请先设置 RAGQA_LLAMA_MODEL}"
python -m vllm.entrypoints.openai.api_server \
--model "$RAGQA_LLAMA_MODEL" \
--tensor_parallel_size "${RAGQA_TP_SIZE:-2}" \
--gpu_memory_utilization 0.7 \
--port 50001 --max-num-seqs 256 --max-model-len 4096 \
--use-v2-block-manager --disable-log-requests