Why is the weight file of the 6.7b model as large as 27Gb?
#6
by
wangdafa
- opened
6 Safetensors files should be 27GB, but as far as I know, the weight file of Deepseek-Coder-6.7B should be only about 13.5GB. Why?
Hi @wangdaf, the reason is that our uploaded model were with fp32 safetensors default, so it’s 2x larger than DeepSeek-Coder-6.7B which used bf16 safetensors
wangdafa
changed discussion status to
closed