suv11235/olmOCR-7B-grpo-v3
Partial fine-tune vs allenai/olmOCR-2-7B-1025. Partial change: 87% of the weights changed (attention, embeddings, mlp, norms), in 32 of 32 layers.
Tensors changed: Embeddings 1/2, Norms 40/122, MLP 261/280, Attention 294/324, lm_head 0/1
2 derivatives checked
- mradermacher/olmOCR-7B-grpo-v3-i1-GGUF: Full fine-tune (fewer layers)
- mradermacher/olmOCR-7B-grpo-v3-GGUF: Partial fine-tune (fewer layers)