Tag: fine-tuning

All the articles with the tag "fine-tuning".

• Fine-tuning Qwen2.5-1.5B for function calling with QLoRA: the dataset didn't match the format I needed (Part 2)

  7 Sep, 2026

  Choosing a function-calling dataset turned out to be the easy part. Converting it into a format Qwen's chat template actually accepts, and getting loss masking right, took most of the work.

• Fine-tuning Qwen2.5-1.5B for function calling with QLoRA: what the baseline already gets right (Part 1)

  6 Sep, 2026

  Before fine-tuning Qwen2.5-1.5B-Instruct for function calling, I tested how good the untrained baseline already is. The answer changes what this project actually needs to prove.