Test 3

#26
by muellerzr HF staff - opened
Hugging Face Internal Testing Organization org
edited Aug 22, 2023

Model Memory Requirements

These calculations were measured from the Model Memory Utility Space on the Hub.

The minimum recommended vRAM needed for this model to perform inference via Accelerate or device_map="auto" is denoted by the size of the "largest layer" and training of the model is roughly 4x its total size (for Adam).

Results

dtype Largest Layer Total Size Training using Adam
float32 31.29 MB 31.61 MB 126.46 MB
float16 15.64 MB 15.81 MB 63.23 MB
int8 7.82 MB 7.9 MB 31.61 MB
int4 3.91 MB 3.95 MB 15.81 MB
muellerzr changed discussion title from [AUTOMATED] Model Memory Requirements to Test 3

Sign up or log in to comment