Always run SiLU activation in float32 for LLaMA and Mistral #1540
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
abuelnasr0
pushed a commit
to abuelnasr0/keras-nlp
that referenced
this pull request
Apr 2, 2024
…am#1540) * Fix discrepency between HF LLaMA and our implementation * Fix Mistral transformer decoder
SamanehSaadat
pushed a commit
to SamanehSaadat/keras-nlp
that referenced
this pull request
Apr 10, 2024
…am#1540) * Fix discrepency between HF LLaMA and our implementation * Fix Mistral transformer decoder
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Outdated suggestions cannot be applied.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.
PyTorch's SiLU always runs in
float32. Running in half-precision causes catastrophic cancellation and leads to huge errors. This PR fixes this issue for both LLaMA and Mistral.Here's the PyTorch implementations:
CPU Kernel: https://github.com/pytorch/pytorch/blob/35c493f2cf9b623bfdc7e6b34dc1cb39690a7919/aten/src/ATen/native/cpu/Activation.cpp#L1221-L1235
CUDA Kernel: https://github.com/pytorch/pytorch/blob/35c493f2cf9b623bfdc7e6b34dc1cb39690a7919/aten/src/ATen/native/cuda/ActivationSiluKernel.cu
Colab verifying this behavior: https://colab.research.google.com/drive/1v5CNVkWJtyIcQVbh-f51GKbqvrvfDyVd?usp=sharing