LoRA Summary¶

The documentation site keeps a simplified and stable summary for the LoRA experiments instead of listing every rank sweep fluctuation.

3B Model¶

Current retained conclusion:

Qwen2.5-3B -> rank 2

Qwen2.5 3B LoRA adaptation

Qwen2.5 3B FastAPI serving

7B Model¶

Current retained conclusion:

Qwen2.5-7B -> rank 4

These conclusions are kept as the current documentation baseline. The experiments themselves still contain randomness, so the site keeps one stable wording for later explanation and reconstruction.