Why is the NVFP4 quantized version larger than the original?

#2
by NexusITComp - opened

Hello everyone!

The original DeepSeek v4 Pro model weighs 865 GB,
while DeepSeek-V4-Pro-NVFP4 weighs 913 GB.
Why is this the case? Where is the savings?

Hello everyone!

The original DeepSeek v4 Pro model weighs 865 GB,
while DeepSeek-V4-Pro-NVFP4 weighs 913 GB.
Why is this the case? Where is the savings?

... regarding that this version is NVFP4 while the original is FP8...

Sign up or log in to comment