pmetal fuse

Merge LoRA adapter weights into the base model weights, producing a standalone model without adapter overhead.

Usage

pmetal fuse \
  --model <BASE_MODEL> \
  --lora <LORA_WEIGHTS> \
  [OPTIONS]

Examples

# Fuse LoRA weights
pmetal fuse \
  --model Qwen/Qwen3-0.6B \
  --lora ./output/lora_weights.safetensors

# Accurate mode (higher precision)
pmetal fuse \
  --model Qwen/Qwen3-0.6B \
  --lora ./output/lora_weights.safetensors \
  --accurate

Modes

Mode	Description
Standard	Fast fusing with standard precision
Accurate	Higher precision fusing for sensitive weights

See Also

• pmetal train — Generate LoRA weights
• pmetal quantize — Quantize the fused model