| # [ragqa] ①监督式查询分解服务（LLaMA-Factory OpenAI 兼容 API，端口 50001） | |
| # 需要环境变量：RAGQA_LLAMA_MODEL（基座）、RAGQA_LLAMA_LORA_SUPERVISED（监督式 LoRA） | |
| : "${RAGQA_LLAMA_MODEL:?请设置 RAGQA_LLAMA_MODEL}" | |
| : "${RAGQA_LLAMA_LORA_SUPERVISED:?请设置 RAGQA_LLAMA_LORA_SUPERVISED}" | |
| HERE="$(cd "$(dirname "$0")" && pwd)" | |
| YAML="$HERE/strategy.runtime.yaml" | |
| cat > "$YAML" <<YML | |
| model_name_or_path: $RAGQA_LLAMA_MODEL | |
| adapter_name_or_path: $RAGQA_LLAMA_LORA_SUPERVISED | |
| template: llama3 | |
| finetuning_type: lora | |
| YML | |
| API_PORT=50001 CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES:-0} llamafactory-cli api "$YAML" | |