liger-kernel-nightly 0.5.10.dev20250609223356__py3-none-any.whl → 0.5.10.dev20250610174206__py3-none-any.whl

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.
@@ -7,6 +7,7 @@ import torch
7
7
 
8
8
  from torch.nn import CrossEntropyLoss
9
9
  from transformers.models.llava.modeling_llava import LlavaCausalLMOutputWithPast
10
+ from transformers.utils import is_torchdynamo_compiling
10
11
 
11
12
  from liger_kernel.transformers.fused_linear_cross_entropy import LigerFusedLinearCrossEntropyLoss
12
13
  from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
@@ -193,6 +194,7 @@ def lce_forward_deprecated(
193
194
  image_hidden_states=image_features if pixel_values is not None else None,
194
195
  )
195
196
 
197
+
196
198
  def lce_forward(
197
199
  self,
198
200
  input_ids: torch.LongTensor = None,
@@ -316,7 +318,6 @@ def lce_forward(
316
318
  logits=logits, labels=labels, vocab_size=self.config.text_config.vocab_size, **lm_kwargs
317
319
  )
318
320
 
319
-
320
321
  if not return_dict:
321
322
  output = (logits,) + outputs[1:]
322
323
  return (loss,) + output if loss is not None else output
@@ -316,7 +316,7 @@ def apply_liger_kernel_to_llava(
316
316
  if fused_linear_cross_entropy:
317
317
  if transformer_version >= version.parse("4.52.0"):
318
318
  modeling_llava.LlavaForConditionalGeneration.forward = llava_lce_forward
319
- elif transformer_version >= version.parse("4.49.0") and transformer_version < version.parse("4.52.0"):
319
+ elif transformer_version >= version.parse("4.49.0") and transformer_version < version.parse("4.52.0"):
320
320
  modeling_llava.LlavaForConditionalGeneration.forward = llava_lce_forward_deprecated
321
321
  else: # if version < 4.49.0
322
322
  logger.warning(
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: liger_kernel_nightly
3
- Version: 0.5.10.dev20250609223356
3
+ Version: 0.5.10.dev20250610174206
4
4
  Summary: Efficient Triton kernels for LLM Training
5
5
  License: BSD 2-CLAUSE LICENSE
6
6
  Copyright 2024 LinkedIn Corporation
@@ -53,7 +53,7 @@ liger_kernel/transformers/grpo_loss.py,sha256=uAkUNKSnUGEOqa82L9w2e6AI1kcmG8K45-
53
53
  liger_kernel/transformers/jsd.py,sha256=DGqRnxIZxsvxo0_tbbxX3b-sDbDjC_yKufyRIHCcScY,2979
54
54
  liger_kernel/transformers/kl_div.py,sha256=WLffFbh1EExD2Eb1F7lN11fo9JJC-0751WJjZAF1Fj8,409
55
55
  liger_kernel/transformers/layer_norm.py,sha256=c9pk3PEasOKYR0rhe5e5nNrnYKVCEW4VC8S6LpCq9EQ,906
56
- liger_kernel/transformers/monkey_patch.py,sha256=Z9HcewGGKPruHU9NHfocV4vvdj13TuIfhmBnuNw1hk0,74606
56
+ liger_kernel/transformers/monkey_patch.py,sha256=zeqmbU__X965iSZ4ZO0Zq3kq6qvqfSgU7B3acxynL3Y,74605
57
57
  liger_kernel/transformers/multi_token_attention.py,sha256=l9VDICK0dfmifUDW668hGscP8AHq2rYcM2oGUa3baRQ,1751
58
58
  liger_kernel/transformers/qwen2vl_mrope.py,sha256=5EwSqrMdsL9MYspeBMXBsNJKvH0MOmRrtJXAJlnnlOI,1047
59
59
  liger_kernel/transformers/rms_norm.py,sha256=eErIr1n-13oVrc1VJY07lqazYelw_vlu9Az__RmXPSE,2717
@@ -70,7 +70,7 @@ liger_kernel/transformers/model/gemma2.py,sha256=ORmzklEAMpk93nToRo4d_ZJbM4ScVE2
70
70
  liger_kernel/transformers/model/gemma3.py,sha256=JI4jj9K660HeRsofB6cpkCHBQ0OsazElArRtKUehUmw,15945
71
71
  liger_kernel/transformers/model/glm4.py,sha256=GlnEhdGJuDIqp2R9qC54biY3HwV1tWmfpJm6ijoAsrM,5257
72
72
  liger_kernel/transformers/model/llama.py,sha256=LcIxVfF0PXXWHBVJa6Ody_5fAtIpxQcI4jC_j-o51fU,12503
73
- liger_kernel/transformers/model/llava.py,sha256=hzPeBfDBbgu9UXMdbjgFRaKYKZZaNb4FbTngn6bLcGg,15020
73
+ liger_kernel/transformers/model/llava.py,sha256=bLCioday_SOm69ogMDBhy_4UsVkH2-BSl93-EXY6-7I,15076
74
74
  liger_kernel/transformers/model/loss_utils.py,sha256=WWAMdiONPaXpIvxyOim_0igLrYh0yyOok5Q9_L9xvZw,1787
75
75
  liger_kernel/transformers/model/mistral.py,sha256=okKkyashfFLfhjIT--f3JY6JHOslOtDI8U1dlpBC2Zs,5565
76
76
  liger_kernel/transformers/model/mixtral.py,sha256=VY-y73IyjcCyWyI7ahxXLw0fJrhgjYfr1xwRYtsHX0o,11396
@@ -87,9 +87,9 @@ liger_kernel/transformers/trainer/__init__.py,sha256=p7yQfklV8-467qSz_ZMimkbDF7H
87
87
  liger_kernel/transformers/trainer/orpo_trainer.py,sha256=tX0h63aOFe3rNqTmk6JpMf75UPo981yzEa6TghnjS0Q,5370
88
88
  liger_kernel/triton/__init__.py,sha256=qCiCamzCRv6lpV8IqpAc9YMdNKC7GKurClWceQPnlis,92
89
89
  liger_kernel/triton/monkey_patch.py,sha256=Rd0hUHAzDkFfHvnX7-PBaNK5EKnZhtfM_h-fgQH9HPY,1568
90
- liger_kernel_nightly-0.5.10.dev20250609223356.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
91
- liger_kernel_nightly-0.5.10.dev20250609223356.dist-info/METADATA,sha256=wn7QobLSoNLSi0rDXabRP1Rn5s5hXnyToI11iH-edq0,24309
92
- liger_kernel_nightly-0.5.10.dev20250609223356.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
93
- liger_kernel_nightly-0.5.10.dev20250609223356.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
94
- liger_kernel_nightly-0.5.10.dev20250609223356.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
95
- liger_kernel_nightly-0.5.10.dev20250609223356.dist-info/RECORD,,
90
+ liger_kernel_nightly-0.5.10.dev20250610174206.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
91
+ liger_kernel_nightly-0.5.10.dev20250610174206.dist-info/METADATA,sha256=T6CCI8j-_GLD4_OTFov5VFLiGK7sITnt6Ht6zVDPhqw,24309
92
+ liger_kernel_nightly-0.5.10.dev20250610174206.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
93
+ liger_kernel_nightly-0.5.10.dev20250610174206.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
94
+ liger_kernel_nightly-0.5.10.dev20250610174206.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
95
+ liger_kernel_nightly-0.5.10.dev20250610174206.dist-info/RECORD,,