Remove seq_len arg in rotary_emb #1443
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Remove
seq_lenargument inself.rotary_emb()in the flash-attn and xformers patches. Solves #1423.Description
Issue #1423 mentioned that the error
TypeError: LlamaRotaryEmbedding.forward() got an unexpected keyword argument 'seq_len'occurs when training Llama with flash-attention. I found that it only happens when the version of transformers is greater than 4.38. This error occurs because in transformers version 4.39, theseq_lenargument in theLlamaRotaryEmbedding.forward()function was deprecated. You can see the change at huggingface/transformers@ffe60fd#diff-06392bad3b9e97be9ade60d4ac46f73b6809388f4d507c2ba1384ab872711c51.In the file
axolotl/src/axolotl/monkeypatch/llama_attn_hijack_flash.py, line 290 callsLlamaRotaryEmbedding.forward()with the argumentseq_len=kv_seq_len, which is no longer supported in the updated version of transformers.Motivation and Context
This change solves issue #1423, allowing flash-attn to be enabled when training with Llama.
How has this been tested?
The changes are small. I re-ran my training script after removing the
seq_lenrelated arguments and observed the same behavior as when downgrading transformers to version 4.38.