prithivMLmods/DeepSeek-OCR-Latest-BF16.I64

Same weights, repacked vs deepseek-ai/DeepSeek-OCR. Same weights in different files (re-saved, re-sharded or new metadata), plus 1 new tensors.

Tensors changed: lm_head 0/1, Embeddings 0/7, other 0/12, Norms 0/171, MLP 0/2303, Attention 0/216

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse