jaqja/HunYuanOCR-SFT

Full fine-tune vs tencent/HunyuanOCR. Fine-tune: 100% of the weights changed, across 27 of 27 layers.

Tensors changed: Embeddings 4/4, Norms 205/205, MLP 182/182, Attention 312/312, other 10/10

2 derivatives checked

• mradermacher/HunYuanOCR-SFT-i1-GGUF: Quantization (fewer layers)
• mradermacher/HunYuanOCR-SFT-GGUF: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse