liger-kernel-nightly 0.5.2.dev20250101082227__py3-none-any.whl → 0.5.2.dev20250108072837__py3-none-any.whl

liger_kernel/chunked_loss/cpo_loss.py

@@ -33,7 +33,11 @@ class LigerFusedLinearCPOFunction(LigerFusedLinearPreferenceBase):
         loss = (-F.logsigmoid(logits) * (1 - label_smoothing) - F.logsigmoid(-logits) * label_smoothing).sum() / (
             full_target.shape[0] // 2
         )
-        return loss
+
+        chosen_rewards = beta * chosen_logps
+        rejected_rewards = beta * rejected_logps
+
+        return loss, chosen_rewards, rejected_rewards
 
     @staticmethod
     def forward(

liger_kernel/chunked_loss/simpo_loss.py

@@ -42,7 +42,10 @@ class LigerFusedLinearSimPOFunction(LigerFusedLinearPreferenceBase):
             full_target.shape[0] // 2
         )
 
-        return loss
+        chosen_rewards = beta * chosen_logps
+        rejected_rewards = beta * rejected_logps
+
+        return loss, chosen_rewards, rejected_rewards
 
     @staticmethod
     def forward(

liger_kernel/ops/cross_entropy.py

@@ -95,7 +95,8 @@ def liger_cross_entropy_kernel(
         return
 
     loss_ptr += program_id * loss_stride
-    z_loss_ptr += program_id * loss_stride
+    if RETURN_Z_LOSS == _TRUE:
+        z_loss_ptr += program_id * loss_stride
 
     if HAS_WEIGHT:
         weight_y = tl.load(weight_ptr + y).cast(tl.float32)
@@ -296,7 +297,7 @@ def cross_entropy_forward(
     if return_z_loss == _TRUE.value:
         z_loss_1d = torch.zeros(n_rows, dtype=_input.dtype, device=_input.device)
     else:
-        z_loss_1d = loss_1d  # dummy ptr when return_z_loss == False
+        z_loss_1d = None  # set None when return_z_loss == False
 
     target_mask = target != ignore_index
     n_non_ignore = target_mask.sum().item()

{liger_kernel_nightly-0.5.2.dev20250101082227.dist-info → liger_kernel_nightly-0.5.2.dev20250108072837.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.5.2.dev20250101082227
+Version: 0.5.2.dev20250108072837
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 

{liger_kernel_nightly-0.5.2.dev20250101082227.dist-info → liger_kernel_nightly-0.5.2.dev20250108072837.dist-info}/RECORD

@@ -3,15 +3,15 @@ liger_kernel/env_report.py,sha256=uhdEC8OydxoZlb7B6YYcAaBF3crGFdIck-4cxaW4NJY,17
 liger_kernel/utils.py,sha256=HJa-xVKOohDn6pLVIx-Fv0V9h0QAL3qZGQNRICI-OpI,249
 liger_kernel/chunked_loss/README.md,sha256=K6rucm6nqHpWCmxUOhBYcE3apwQxAy0TfRUippR7Icw,2243
 liger_kernel/chunked_loss/__init__.py,sha256=R2wCcz4Y0kTAve926DH3k182XKezpXeACMHj05g9Mm8,346
-liger_kernel/chunked_loss/cpo_loss.py,sha256=L4Nk38Xh5Yfhah3Vsc_sN_Q75FWt1LA-xNNXzsK8iPM,3516
+liger_kernel/chunked_loss/cpo_loss.py,sha256=MCR4TzuBoJEaU0IJ7dIreLacQeXLKETV5CegNjhCD9M,3646
 liger_kernel/chunked_loss/dpo_loss.py,sha256=VYZMOafdvE8xlhvTtwjrz81tIzxR1mHF4lXdsADnIQg,4373
 liger_kernel/chunked_loss/functional.py,sha256=9Gr-YXIuEzEJkBUhDx3G2fuQayckLor7cC7svhmPML4,549
 liger_kernel/chunked_loss/fused_linear_distillation.py,sha256=uQtwtu-kaUZJTjNhAnIr3O794oUlUZ98XR5shYtwP5k,10440
 liger_kernel/chunked_loss/fused_linear_preference.py,sha256=25sTgvphLKAR0jyJcrsJPKK1abFpTKrajSyAx8nJ3bc,16134
 liger_kernel/chunked_loss/orpo_loss.py,sha256=jbZxx-EjPK71A6CSyNzTOAIEQgAUjfvwSViw6R_pPXQ,3510
-liger_kernel/chunked_loss/simpo_loss.py,sha256=ZvDIjT9EQrbwzH2LNZMhv84SPsOHGi_Ywk95vgA0b_o,3736
+liger_kernel/chunked_loss/simpo_loss.py,sha256=3TTc7U79Orjgi-Wu81WZkWk5MgsdqKXIOBHgIvDazPw,3865
 liger_kernel/ops/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-liger_kernel/ops/cross_entropy.py,sha256=4zSPzdPl-d2tB3ZOj7uRMpzI4RzZMNLUzkh6eMkH5kU,19179
+liger_kernel/ops/cross_entropy.py,sha256=zi2xsa8ky7M1vySUAGjXMQDFQFkKmGQV-myRIIQM13M,19210
 liger_kernel/ops/fused_linear_cross_entropy.py,sha256=j7cgR95rFAwtPsWZ00PfMwis5F7dtO3EVEw0rZ1GPJk,10231
 liger_kernel/ops/fused_linear_jsd.py,sha256=eKqaADj7LgWfoYqyH03tjrmhNTfJOF1Dhx_bWzBTnTU,9600
 liger_kernel/ops/geglu.py,sha256=axGvCIvlBzuluoAIrWTsp2iZM4BFKNInkPov8YVvH9E,4126
@@ -58,9 +58,9 @@ liger_kernel/transformers/trainer/__init__.py,sha256=p7yQfklV8-467qSz_ZMimkbDF7H
 liger_kernel/transformers/trainer/orpo_trainer.py,sha256=MId1S_MfA3pPVQA1rkiKxp-jZDNz8VmvZzXC-Kugol4,7662
 liger_kernel/triton/__init__.py,sha256=qCiCamzCRv6lpV8IqpAc9YMdNKC7GKurClWceQPnlis,92
 liger_kernel/triton/monkey_patch.py,sha256=Rd0hUHAzDkFfHvnX7-PBaNK5EKnZhtfM_h-fgQH9HPY,1568
-liger_kernel_nightly-0.5.2.dev20250101082227.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
-liger_kernel_nightly-0.5.2.dev20250101082227.dist-info/METADATA,sha256=gNuR5mtVV7fQsT0qPLr3_Ok2WLKHgbC2FidkcY1q6OA,21055
-liger_kernel_nightly-0.5.2.dev20250101082227.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
-liger_kernel_nightly-0.5.2.dev20250101082227.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-liger_kernel_nightly-0.5.2.dev20250101082227.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
-liger_kernel_nightly-0.5.2.dev20250101082227.dist-info/RECORD,,
+liger_kernel_nightly-0.5.2.dev20250108072837.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
+liger_kernel_nightly-0.5.2.dev20250108072837.dist-info/METADATA,sha256=HwmQEBRYnwwbdkzuW53_qsmTSSbi8qu20cVOHsq6B_s,21055
+liger_kernel_nightly-0.5.2.dev20250108072837.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
+liger_kernel_nightly-0.5.2.dev20250108072837.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+liger_kernel_nightly-0.5.2.dev20250108072837.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
+liger_kernel_nightly-0.5.2.dev20250108072837.dist-info/RECORD,,