Add support of Falcon-H1 models for liger kernels #874
+76
−0
Conversation
This file contains hidden or 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
This PR enables ROPE, rmsnorm and cross entropy liger kernels for Falcon-H1 models. Support for fused cross entropy and swiglu to be added later.
shimizust
reviewed
Sep 21, 2025
| "smollm3": apply_liger_kernel_to_smollm3, | ||
| "phi3": apply_liger_kernel_to_phi3, | ||
| "paligemma": apply_liger_kernel_to_paligemma, | ||
| "falcon_h1": apply_liger_kernel_to_falcon_h1, |
There was a problem hiding this comment.
Can you add this to src/liger_kernel/transformers/init.py?
| _patch_rms_norm_module(decoder_layer.post_mlp_layernorm) | ||
|
|
||
|
|
||
| def apply_liger_kernel_to_falcon_h1( |
There was a problem hiding this comment.
Can you add corresponding convergence and monkey patch tests?
Comment on lines
+1943
to
+1950
| cross_entropy (bool): Whether to apply Liger's cross entropy loss. Default is False. | ||
| fused_linear_cross_entropy (bool): | ||
| Whether to apply Liger's fused linear cross entropy loss. Default is True. | ||
| `cross_entropy` and `fused_linear_cross_entropy` cannot both be True. | ||
| If `fused_linear_cross_entropy` is True, the logits will not be materialized but more memory efficient. | ||
| rms_norm (bool): Whether to apply Liger's RMSNorm. Default is True. | ||
| swiglu (bool): Whether to apply Liger's SwiGLU MLP. Default is True. | ||
| model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been |
There was a problem hiding this comment.
Set correct defaults for flce and swiglu if not implemented
Comment on lines
+1967
to
+1987
| if swiglu: | ||
| modeling_falcon_h1.FalconH1MLP = LigerSwiGLUMLP | ||
|
|
||
| if cross_entropy: | ||
| if transformer_version >= version.parse(SUPPORTED_TRANSFORMER_VERSION): | ||
| logger.info("Apply liger cross entropy") | ||
| from transformers.loss.loss_utils import nn | ||
|
|
||
| nn.functional.cross_entropy = liger_cross_entropy | ||
| else: | ||
| logger.warning(TRANSFORMER_DEPRECATION_WARNING) | ||
| modeling_falcon_h1.CrossEntropyLoss = LigerCrossEntropyLoss | ||
|
|
||
| # TODO: To be enabled | ||
| if fused_linear_cross_entropy: | ||
| if transformer_version >= version.parse(SUPPORTED_TRANSFORMER_VERSION): | ||
| modeling_falcon_h1.FalconH1ForCausalLM.forward = llama_lce_forward | ||
| else: # if version < 4.46.1 | ||
| logger.warning(TRANSFORMER_DEPRECATION_WARNING) | ||
| modeling_falcon_h1.FalconH1ForCausalLM.forward = llama_lce_forward_deprecated | ||
|
|
There was a problem hiding this comment.
Can you raise NotImplementedError in the case of swiglu and flce are set to true? Right now we're patching with incorrect modules (also in the case below where model instance exists)
|
Thanks for all the comments @shimizust . Will address soon |
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.
Summary
This PR enables ROPE, rmsnorm and cross entropy liger kernels for Falcon-H1 models
Falcon H1 models -
https://huggingface.co/collections/tiiuae/falcon-h1-6819f2795bc406da60fab8df
Support for fused cross entropy and swiglu to be added later.
Testing Done
Verified fine tuning Falcon H1 models with and without enabling liger kernels and loss plot matches. And gives speedup as well as memory savings.
make testto ensure correctnessmake checkstyleto ensure code stylemake test-convergenceto ensure convergence