Files
llm-gguf-quant-template/docs/WORKFLOW_TEMPLATE.md
2026-03-02 23:22:33 +08:00

1.7 KiB
Raw Blame History

Workflow Template

本流程用于新模型接入,默认在仓库根目录执行。

Step 1: HF -> BF16 GGUF

使用 ik_llama.cpp 的转换脚本:

python convert_hf_to_gguf.py \
  <hf_model_dir> \
  --outtype bf16 \
  --outfile artifacts/<model_name>/base_gguf/<model_name>-bf16.gguf

Step 2: 准备校准数据

./.venv/bin/python scripts/prepare_calib_data.py --force-refresh

输出:

  • calibration/calibration_data_v5_rc.txt
  • calibration/calibration_data_v5_rc_code.txt

固定组成1152 + 2000 + 1000 = 4152 blocks。

Step 3: 生成 imatrix

docker run --gpus all --rm \
  --entrypoint sh \
  -v <repo_root>:/workspace/models \
  -v <repo_root>/calibration/calibration_data_v5_rc_code.txt:/workspace/calib_data.txt \
  hotwa/ik:latest \
  -c "/llama-imatrix -m <bf16_gguf> -f /workspace/calib_data.txt -o <imatrix_out> --ctx-size 512 -ngl 99 --threads 16"

Step 4: 量化导出

分别执行:

docker run --gpus all --rm \
  --entrypoint sh \
  -v <repo_root>:/workspace/models \
  hotwa/ik:latest \
  -c "/llama-quantize --imatrix <imatrix_out> <bf16_gguf> <out_gguf> IQ4_KS"

将量化结果放入:artifacts/<model_name>/quantized_gguf/

Step 5: 组织上传目录

cp templates/modelscope/README.template.md modelscope_upload/README.md
cp templates/modelscope/configuration.template.json modelscope_upload/configuration.json
cp templates/modelscope/.gitattributes modelscope_upload/.gitattributes

然后把目标发布文件复制到 modelscope_upload/

Step 6: 上传

./scripts/upload_to_modelscope.sh <repo_id> <token> modelscope_upload direct "Upload quantized GGUF"
  • direct:关闭代理上传
  • proxy:保留代理上传