formalmathatepfl/qwen3-cpt
Full fine-tune vs Qwen/Qwen3-8B-Base. Fine-tune: 92% of the weights changed, across 36 of 36 layers.
Tensors changed: Embeddings 0/1, Norms 124/145, MLP 108/108, Attention 144/144, lm_head 1/1
7 derivatives checked
- formalmathatepfl/qwen3-8b-feedback-no-repair: Full fine-tune
- formalmathatepfl/qwen3-sft-feedback-with-proof-repair: Full fine-tune
- formalmathatepfl/qwen3-8b-feedback-with-repair: Full fine-tune
- VladShash/qwen3-8b-classic-with-repair: Full fine-tune
- VladShash/qwen3-8b-classic-no-repair: Full fine-tune
- formalmathatepfl/qwen3-8b-classic-one-shot: Full fine-tune
- etiennebamas/qwen3-sft-classic-small-data-equal-compute: Full fine-tune