BDRC/tibetan-ocr

Full fine-tune vs PaddlePaddle/PaddleOCR-VL-1.6. Fine-tune: 100% of the weights changed, across 27 of 27 layers.

Tensors changed: MLP 165/166, Norms 91/149, Attention 288/288, Embeddings 3/3

1 derivatives checked

• nakamura196/yigdzin1-gguf: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse