Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.benchgen.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Merge a LoRA Adapter

After a fine-tuning run completes, you have a LoRA adapter — a small set of weight deltas. Merging combines the adapter with the base model into a single, self-contained checkpoint you can deploy or evaluate.

​
When to Merge

Merge when:
  • The fine-tuning run completed successfully and the loss curve converged.
  • You want to deploy the model inside an Agent or evaluate it in Eval.
  • You want a portable checkpoint (merged models can run without the adapter separately).
You don’t need to merge to test — Train provides an inference endpoint against the adapter directly (see Run inference).

​
Steps

  1. Go to Train → Runs and open the completed run.
  2. Click Merge adapter.
  3. Choose the output format:
    • safetensors (default, recommended)
    • GGUF (for local/edge deployment)
  4. Click Start merge — this typically takes 2–5 minutes.
  5. When complete, the merged model appears in Train → Models and in Eval → Models.

​
Next Steps