trillionlabs/GravityOCR
Full fine-tune vs zai-org/GLM-OCR. Fine-tune: 100% of the weights changed, across 24 of 24 layers.
Tensors changed: Norms 80/164, MLP 179/179, Attention 160/160, other 3/3, Embeddings 2/2
Full fine-tune vs zai-org/GLM-OCR. Fine-tune: 100% of the weights changed, across 24 of 24 layers.
Tensors changed: Norms 80/164, MLP 179/179, Attention 160/160, other 3/3, Embeddings 2/2