zr-wang/AgenticOCR-8B
Partial fine-tune vs Qwen/Qwen3-VL-8B-Instruct. Partial change: 88% of the weights changed (attention, embeddings, lm_head, mlp, norms), in 36 of 36 layers.
Tensors changed: Embeddings 1/4, Norms 127/261, MLP 120/232, Attention 144/252, lm_head 1/1
1 derivatives checked
- mradermacher/AgenticOCR-8B-GGUF: Same weights, converted (fewer layers)