Fix errors introduced by GPTQ UX

neuralmagic / sparseml

Libraries for applying sparsification recipes to neural networks with a few lines of code, enabling faster and smaller models

Apache License 2.0

2.01k stars 140 forks source link

Closed rahul-tuli closed 1 month ago

rahul-tuli commented 1 month ago

This PR makes main green again by:

1) Install compressed tensors from source wherever transformers deps are needed 2) Apply a commit that was missed (for transformers tests)

Satrat commented 1 month ago

Could you add one more change to that PR:

from src.sparseml.modifiers.quantization.gptq.utils.gptq_wrapper import GPTQWrapper

should be

from sparseml.modifiers.quantization.gptq.utils.gptq_wrapper import GPTQWrapper

otherwise we won't be able to run anything outside of sparseML root directory, this was causing my 7b example runs to fail