LoRA Adapter Merge¶

After LoRA training, you can merge the adapter into the base model and export a full HuggingFace model directory.

Script¶

scripts/merge_lora_adapter.sh

Usage: adapter directory¶

Use this when the adapter path is a directory containing adapter_model.safetensors and adapter_config.json.

scripts/merge_lora_adapter.sh \
  --base_model Qwen/Qwen2.5-7B-Instruct \
  --adapter_path /path/to/lora_adapter \
  --output_dir /path/to/merged_hf_model

Usage: single adapter file¶

Use this when the adapter is a single safetensors file.

scripts/merge_lora_adapter.sh \
  --base_model Qwen/Qwen2.5-7B-Instruct \
  --adapter_path /path/to/adapter.safetensors \
  --adapter_config /path/to/adapter_config.json \
  --output_dir /path/to/merged_hf_model

The output directory is a standard HuggingFace model folder that can be loaded with from_pretrained.