liger-kernel-nightly 0.4.1.dev20241115210858__tar.gz → 0.4.2.dev20241117192137__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {liger_kernel_nightly-0.4.1.dev20241115210858/src/liger_kernel_nightly.egg-info → liger_kernel_nightly-0.4.2.dev20241117192137}/PKG-INFO +1 -1
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/pyproject.toml +1 -1
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/monkey_patch.py +5 -2
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137/src/liger_kernel_nightly.egg-info}/PKG-INFO +1 -1
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/LICENSE +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/NOTICE +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/README.md +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/setup.cfg +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/chunked_loss/__init__.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/chunked_loss/dpo_loss.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/chunked_loss/fused_linear_preference.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/chunked_loss/orpo_loss.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/env_report.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/__init__.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/cross_entropy.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/experimental/embedding.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/experimental/mm_int8int2.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/fused_linear_cross_entropy.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/fused_linear_jsd.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/geglu.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/group_norm.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/jsd.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/kl_div.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/layer_norm.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/rms_norm.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/rope.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/swiglu.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/ops/utils.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/__init__.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/auto_model.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/cross_entropy.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/experimental/embedding.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/functional.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/fused_linear_cross_entropy.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/fused_linear_jsd.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/geglu.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/group_norm.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/jsd.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/kl_div.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/layer_norm.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/__init__.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/gemma.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/gemma2.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/llama.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/mistral.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/mixtral.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/mllama.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/phi3.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/qwen2.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/model/qwen2_vl.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/rms_norm.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/rope.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/swiglu.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/transformers/trainer_integration.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/triton/__init__.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel/triton/monkey_patch.py +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel_nightly.egg-info/SOURCES.txt +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel_nightly.egg-info/dependency_links.txt +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel_nightly.egg-info/requires.txt +0 -0
- {liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/src/liger_kernel_nightly.egg-info/top_level.txt +0 -0
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
4
4
|
|
5
5
|
[project]
|
6
6
|
name = "liger_kernel_nightly"
|
7
|
-
version = "0.4.
|
7
|
+
version = "0.4.2.dev20241117192137"
|
8
8
|
description = "Efficient Triton kernels for LLM Training"
|
9
9
|
urls = { "Homepage" = "https://github.com/linkedin/Liger-Kernel" }
|
10
10
|
readme = { file = "README.md", content-type = "text/markdown" }
|
@@ -56,12 +56,15 @@ def _bind_method_to_module(module, method_name: str, new_method: Callable):
|
|
56
56
|
module.__dict__[method_name] = new_method.__get__(module, module.__class__)
|
57
57
|
|
58
58
|
|
59
|
-
def _patch_rms_norm_module(
|
59
|
+
def _patch_rms_norm_module(
|
60
|
+
module, offset=0.0, eps=1e-6, casting_mode="llama", in_place=True
|
61
|
+
):
|
60
62
|
module.offset = offset
|
61
63
|
module.casting_mode = casting_mode
|
62
64
|
module.variance_epsilon = (
|
63
65
|
getattr(module, "variance_epsilon", None) or getattr(module, "eps", None) or eps
|
64
66
|
)
|
67
|
+
module.in_place = in_place
|
65
68
|
_bind_method_to_module(module, "forward", LigerRMSNorm.forward)
|
66
69
|
_bind_method_to_module(module, "extra_repr", LigerRMSNorm.extra_repr)
|
67
70
|
|
@@ -510,7 +513,7 @@ def apply_liger_kernel_to_gemma2(
|
|
510
513
|
LigerRMSNorm, offset=1.0, casting_mode="gemma", init_fn="zeros", in_place=False
|
511
514
|
)
|
512
515
|
_patch_rms_norm_module_for_gemma2 = partial(
|
513
|
-
_patch_rms_norm_module, offset=1.0, casting_mode="gemma"
|
516
|
+
_patch_rms_norm_module, offset=1.0, casting_mode="gemma", in_place=False
|
514
517
|
)
|
515
518
|
|
516
519
|
if rope:
|
File without changes
|
{liger_kernel_nightly-0.4.1.dev20241115210858 → liger_kernel_nightly-0.4.2.dev20241117192137}/NOTICE
RENAMED
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|