yujiepan/deepseek-v4-fp4-fp8-tiny-random
Different architecture vs deepseek-ai/DeepSeek-V4-Pro. Different architecture: only 67 of 145116/6441 tensors match by name and shape.
4 other repos hold byte-for-byte identical weight files.
Different architecture vs deepseek-ai/DeepSeek-V4-Pro. Different architecture: only 67 of 145116/6441 tensors match by name and shape.
4 other repos hold byte-for-byte identical weight files.