AWQ: Patch for mismatched devices in RotaryEmbedding #1480
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Describe your changes
In transformers>4.43 which is required for models such as llama-3.1 and phi-3.5, there is an open issue related to device mismatch in the rotary embedding module huggingface/transformers#32420
There is no fix yet in either transformers or autoawq so we patch the model adapter in Olive based on the package versions. This unblocks quantization using newer transformers version for llama like models. Fix is based on casper-hansen/AutoAWQ#630
AutoGPTQ has the same issue but is already fixed in main so users can install from source. Also there is no straigtforward way to do a similar patch with it. Tried adding
model.rotary_embed
tooutside_layer_modules
but it fails withget_device
since the module has no parameters.Checklist before requesting a review
lintrunner -a
(Optional) Issue link