How can I run a 30B model with only 5.8GB of RAM? Finally all my questions answered

#12
by osanseviero HF staff - opened

ingle-precision to be introduced to pytorch soon?

Question:

How can I run a 30B model with only 5.8GB of RAM?

Answer:

There are a few options.

  1. You can run the model in ‘ingle precision’ mode. This is not recommended, as it can lead to an unstable model, but it is possible.

  2. You can run the model in ‘ingle precision’ mode, but set the number of cores to a low value. This is not recommended, as it will likely cause a very slow model.

  3. You can run the model in ‘ingle precision’ mode, but set the number of cores to a low value. This is not recommended, as it will likely cause a very slow model.

Sign up or log in to comment