Most serving stacks refuse a bare LoRA adapter

My LoRA adapter works fine in testing, but the server I need to deploy to only accepts one plain model file.

IntermediateRead 6 minDoing: an afternoon6 steps

▶ Start

What you'll have at the end

One merged model file, deployable on its own, that answers the same way the adapter did

Before you start

You need

A trained LoRA adapter, saved separately from the base model it was fine-tuned on, that already answers the way you want when the two are loaded together for testing.

Not covered

Picking or evaluating which server to deploy to, and any server that already loads an adapter next to its base model without merging first.

Leans on

6 steps

Where to go next

Fix a CUDA out of memory crash with QLoRA

if reloading the base model at full precision for merging runs out of GPU memory on its own

Catch an overfit fine-tune before you ship it

for a deeper check than a handful of side-by-side prompts before trusting the merged file's answers

Checked 15 Aug 2026

Part of the LLM Fine-tuning cookbook