allenai/olmOCR-7B-0225-preview
Full fine-tune vs Qwen/Qwen2-VL-7B-Instruct. Fine-tune: 93% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/2, Norms 140/187, MLP 215/216, Attention 324/324, lm_head 1/1
9 derivatives checked
- bartowski/allenai_olmOCR-7B-0225-preview-GGUF: Quantization (fewer layers)
- mradermacher/olmOCR-7B-0225-preview-i1-GGUF: Quantization (fewer layers)
- mradermacher/olmOCR-7B-0225-preview-GGUF: Same weights, converted (fewer layers)
- lmstudio-community/olmOCR-7B-0225-preview-GGUF: Quantization (fewer layers)
- DevQuasar/allenai.olmOCR-7B-0225-preview-GGUF: Same weights, converted (fewer layers)
- tngtech/olmOCR-7B-faithful: Full fine-tune
- Adun/olmOCR-7B-thai-v3: LoRA-style
- Adun/olmOCR-7B-thai-v2: LoRA-style
- Adun/olmOCR-7B-thai-v3.2: LoRA-style
2 other repos hold byte-for-byte identical weight files.