liger-kernel-nightly 0.5.9.dev20250515065336__tar.gz → 0.5.9.dev20250516193902__tar.gz

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.5.9.dev20250515065336
+Version: 0.5.9.dev20250516193902
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "liger_kernel_nightly"
-version = "0.5.9.dev20250515065336"
+version = "0.5.9.dev20250516193902"
 description = "Efficient Triton kernels for LLM Training"
 urls = { "Homepage" = "https://github.com/linkedin/Liger-Kernel" }
 readme = { file = "README.md", content-type = "text/markdown" }

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/chunked_loss/jsd_loss.py

@@ -150,8 +150,8 @@ class LigerFusedLinearJSDLoss(torch.nn.Module):
         teacher_input: torch.Tensor,
         teacher_weight: torch.Tensor,
         true_labels: torch.LongTensor,
-        student_bias: torch.Tensor,
-        teacher_bias: torch.Tensor,
+        student_bias: torch.Tensor = None,
+        teacher_bias: torch.Tensor = None,
     ) -> torch.Tensor:
         """
         Compute the JSD distillation loss.

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/gemma.py

@@ -8,18 +8,12 @@ import torch
 from torch.nn import CrossEntropyLoss
 from transformers.cache_utils import Cache
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.gemma.modeling_gemma import _CONFIG_FOR_DOC
-from transformers.models.gemma.modeling_gemma import GEMMA_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(GEMMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -129,8 +123,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(GEMMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/gemma2.py

@@ -9,10 +9,6 @@ import torch
 from torch.nn import CrossEntropyLoss
 from transformers.cache_utils import HybridCache
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.gemma2.modeling_gemma2 import _CONFIG_FOR_DOC
-from transformers.models.gemma2.modeling_gemma2 import GEMMA2_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
@@ -136,8 +132,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(GEMMA2_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/gemma3.py

@@ -9,13 +9,9 @@ import torch.nn as nn
 from transformers.cache_utils import Cache
 from transformers.cache_utils import HybridCache
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.gemma3.modeling_gemma3 import _CONFIG_FOR_DOC
-from transformers.models.gemma3.modeling_gemma3 import GEMMA3_INPUTS_DOCSTRING
 from transformers.models.gemma3.modeling_gemma3 import Gemma3CausalLMOutputWithPast
-from transformers.utils import add_start_docstrings_to_model_forward
 from transformers.utils import is_torchdynamo_compiling
 from transformers.utils import logging
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
@@ -25,8 +21,6 @@ logger = logging.get_logger(__name__)
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(GEMMA3_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def causal_forward(
     self,
     input_ids: torch.LongTensor = None,
@@ -141,8 +135,6 @@ def causal_forward(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(GEMMA3_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=Gemma3CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def multimodal_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/glm4.py

@@ -6,18 +6,12 @@ from typing import Union
 import torch
 
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.glm4.modeling_glm4 import _CONFIG_FOR_DOC
-from transformers.models.glm4.modeling_glm4 import GLM4_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(GLM4_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/llama.py

@@ -9,10 +9,6 @@ import torch.nn.functional as F
 
 from torch.nn import CrossEntropyLoss
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.llama.modeling_llama import _CONFIG_FOR_DOC
-from transformers.models.llama.modeling_llama import LLAMA_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
@@ -22,8 +18,6 @@ if TYPE_CHECKING:
     from transformers.cache_utils import Cache
 
 
-@add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -137,8 +131,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/llava.py

@@ -5,19 +5,13 @@ from typing import Union
 
 import torch
 
-from transformers.models.llava.modeling_llava import _CONFIG_FOR_DOC
-from transformers.models.llava.modeling_llava import LLAVA_INPUTS_DOCSTRING
 from transformers.models.llava.modeling_llava import LlavaCausalLMOutputWithPast
-from transformers.utils import add_start_docstrings_to_model_forward
 from transformers.utils import is_torchdynamo_compiling
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
 
 
-@add_start_docstrings_to_model_forward(LLAVA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=LlavaCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -210,9 +204,7 @@ def lce_forward_deprecated(
     )
 
 
-@add_start_docstrings_to_model_forward(LLAVA_INPUTS_DOCSTRING)
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@replace_return_docstrings(output_type=LlavaCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/mistral.py

@@ -7,18 +7,12 @@ import torch
 
 from transformers.cache_utils import Cache
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.mistral.modeling_mistral import _CONFIG_FOR_DOC
-from transformers.models.mistral.modeling_mistral import MISTRAL_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/mixtral.py

@@ -7,19 +7,13 @@ import torch
 
 from torch.nn import CrossEntropyLoss
 from transformers.modeling_outputs import MoeCausalLMOutputWithPast
-from transformers.models.mixtral.modeling_mixtral import _CONFIG_FOR_DOC
-from transformers.models.mixtral.modeling_mixtral import MIXTRAL_INPUTS_DOCSTRING
 from transformers.models.mixtral.modeling_mixtral import load_balancing_loss_func
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(MIXTRAL_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=MoeCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -146,8 +140,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(MIXTRAL_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=MoeCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 # Ignore copy
 def lce_forward(
     self,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/mllama.py

@@ -8,17 +8,12 @@ import torch
 from torch.nn import CrossEntropyLoss
 from transformers.cache_utils import Cache
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.mllama.modeling_mllama import MLLAMA_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(MLLAMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class="MllamaTextConfig")
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -135,8 +130,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(MLLAMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class="MllamaTextConfig")
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/olmo2.py

@@ -6,18 +6,12 @@ from typing import Union
 import torch
 
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.olmo2.modeling_olmo2 import _CONFIG_FOR_DOC
-from transformers.models.olmo2.modeling_olmo2 import OLMO2_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(OLMO2_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/paligemma.py

@@ -7,13 +7,9 @@ import torch
 
 from torch.nn import CrossEntropyLoss
 from transformers.cache_utils import Cache
-from transformers.models.paligemma.modeling_paligemma import _CONFIG_FOR_DOC
-from transformers.models.paligemma.modeling_paligemma import PALIGEMMA_INPUTS_DOCSTRING
 from transformers.models.paligemma.modeling_paligemma import PaliGemmaCausalLMOutputWithPast
-from transformers.utils import add_start_docstrings_to_model_forward
 from transformers.utils import is_torchdynamo_compiling
 from transformers.utils import logging
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
@@ -21,8 +17,6 @@ from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinea
 logger = logging.get_logger(__name__)
 
 
-@add_start_docstrings_to_model_forward(PALIGEMMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=PaliGemmaCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -206,8 +200,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(PALIGEMMA_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=PaliGemmaCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/phi3.py

@@ -7,18 +7,12 @@ import torch
 
 from torch.nn import CrossEntropyLoss
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.phi3.modeling_phi3 import _CONFIG_FOR_DOC
-from transformers.models.phi3.modeling_phi3 import PHI3_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(PHI3_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -128,8 +122,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(PHI3_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/qwen2.py

@@ -7,18 +7,12 @@ import torch
 
 from torch.nn import CrossEntropyLoss
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.qwen2.modeling_qwen2 import _CONFIG_FOR_DOC
-from transformers.models.qwen2.modeling_qwen2 import QWEN2_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 from transformers.utils.deprecation import deprecate_kwarg
 
 from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward_deprecated(
     self,
     input_ids: torch.LongTensor = None,
@@ -127,8 +121,6 @@ def lce_forward_deprecated(
 
 
 @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-@add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/qwen2_5_vl.py

@@ -6,17 +6,11 @@ from typing import Union
 import torch
 
 from torch.nn import CrossEntropyLoss
-from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import _CONFIG_FOR_DOC
-from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import QWEN2_5_VL_INPUTS_DOCSTRING
 from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import Qwen2_5_VLCausalLMOutputWithPast
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(QWEN2_5_VL_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=Qwen2_5_VLCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/qwen2_vl.py

@@ -8,17 +8,11 @@ import torch
 from packaging import version
 from torch.nn import CrossEntropyLoss
 from transformers import __version__ as transformers_version
-from transformers.models.qwen2_vl.modeling_qwen2_vl import _CONFIG_FOR_DOC
-from transformers.models.qwen2_vl.modeling_qwen2_vl import QWEN2_VL_INPUTS_DOCSTRING
 from transformers.models.qwen2_vl.modeling_qwen2_vl import Qwen2VLCausalLMOutputWithPast
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(QWEN2_VL_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=Qwen2VLCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: torch.LongTensor = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/qwen3.py

@@ -5,16 +5,10 @@ from typing import Union
 import torch
 
 from transformers.modeling_outputs import CausalLMOutputWithPast
-from transformers.models.qwen3.modeling_qwen3 import _CONFIG_FOR_DOC
-from transformers.models.qwen3.modeling_qwen3 import QWEN3_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(QWEN3_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: Optional[torch.LongTensor] = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel/transformers/model/qwen3_moe.py

@@ -7,16 +7,10 @@ import torch
 from transformers.modeling_outputs import MoeCausalLMOutputWithPast
 from transformers.modeling_outputs import MoeModelOutputWithPast
 from transformers.models.mixtral.modeling_mixtral import load_balancing_loss_func
-from transformers.models.qwen3_moe.modeling_qwen3_moe import _CONFIG_FOR_DOC
-from transformers.models.qwen3_moe.modeling_qwen3_moe import QWEN3_MOE_INPUTS_DOCSTRING
-from transformers.utils import add_start_docstrings_to_model_forward
-from transformers.utils import replace_return_docstrings
 
 from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
 
 
-@add_start_docstrings_to_model_forward(QWEN3_MOE_INPUTS_DOCSTRING)
-@replace_return_docstrings(output_type=MoeCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
 def lce_forward(
     self,
     input_ids: Optional[torch.LongTensor] = None,

{liger_kernel_nightly-0.5.9.dev20250515065336 → liger_kernel_nightly-0.5.9.dev20250516193902}/src/liger_kernel_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.5.9.dev20250515065336
+Version: 0.5.9.dev20250516193902
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation