NexaAI/DeepSeek-OCR-GGUF-CUDA
Same weights, converted vs deepseek-ai/DeepSeek-OCR. Same weights stored in another dtype (BF16 → F16/F32), plus 33 new tensors.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/25, MLP 0/47, Attention 0/48
Same weights, converted vs deepseek-ai/DeepSeek-OCR. Same weights stored in another dtype (BF16 → F16/F32), plus 33 new tensors.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/25, MLP 0/47, Attention 0/48