zr-wang/AgenticOCR-4B

Full fine-tune vs Qwen/Qwen3-VL-4B-Instruct. Fine-tune: 93% of the weights changed, across 36 of 36 layers.

Tensors changed: Embeddings 2/4, Norms 128/249, MLP 120/220, Attention 144/240

1 derivatives checked

• mradermacher/AgenticOCR-4B-GGUF: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse