Skip to content

Fix quick reference: default recipe is Qwen3-0.6B full fine-tune, not Qwen3.5-2B + LoRA - #179

Merged
alexkroman merged 1 commit into
mainfrom
fix/quick-reference-default-decoder
Nov 17, 2026
Merged

alexkroman merged 1 commit into
mainfrom
fix/quick-reference-default-decoder

Conversation

@alexkroman

Copy link
Copy Markdown
Owner

What this fixes

docs/course/4-quick-reference.md describes the current default recipe (configs/config.yaml + configs/training/production.yaml) as Granite + Qwen3.5-2B, frozen, with LoRA (~80M trained), and lists the freeze_language_model, freeze_text_embed_tokens, and model.use_lora defaults as true. The shipped config says otherwise: text_model_id: Qwen/Qwen3-0.6B, use_lora: false, freeze_language_model: false, freeze_text_embed_tokens: false — the published model's recipe, ~604M trained parameters (596M decoder + 8.4M projector, per the model card).

This corrects the training comment, the Recipes table row, and the three Freeze Flags defaults. Nothing else changes. The course chapters still teach the Qwen3.5-2B frozen recipe as the worked example; whether they should move to the 0.6B recipe is a separate editorial call, left out of this PR on purpose.

The 2B vs 4B question, settled

This audit started from the long-standing flag (Oct 15) that some documents said Qwen3.5-4B while the README and model card said the published model used Qwen3.5-2B. Checking the released weights directly:

  • The only Qwen3.5-based release, revision granite-qwen-frozen-4 of mazesmazes/tiny-audio, ships a config.json whose text_config is qwen3_5_text with hidden_size 2048 and 24 layers, and text_model_id: Qwen/Qwen3.5-2B. Qwen3.5-4B is 2560 hidden x 32 layers. The released decoder is 2B.
  • The 4B exists only as an experiment (issue Move frozen run to Qwen3.5-4B, reinstate Peoples, fix the inference prompt #90, W&B rjid9zj6, 2026-10-16), recorded in configs/training/production.yaml's comments as "kept for a future decoder swap." It was never released.
  • No current document claims 4B; the README and model card's 2B is correct. The current flagship release (Hub main, 2026-10-23) is a different model — Qwen3-0.6B, verified from its released config.json (hidden 1024, 28 layers).

@alexkroman
alexkroman marked this pull request as ready for review November 17, 2026 14:28
@alexkroman
alexkroman merged commit 774f124 into main Nov 17, 2026
5 of 6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant