sgl-project/DeepSeek-V4-Pro-FP8
Different architecture vs deepseek-ai/DeepSeek-V4-Pro. Different architecture: only 2206 of 145116/145054 tensors match by name and shape.
1 other repos hold byte-for-byte identical weight files.
Different architecture vs deepseek-ai/DeepSeek-V4-Pro. Different architecture: only 2206 of 145116/145054 tensors match by name and shape.
1 other repos hold byte-for-byte identical weight files.