liger-kernel-nightly 0.4.2.dev20241119054729__py3-none-any.whl → 0.4.2.dev20241119174706__py3-none-any.whl

liger_kernel/chunked_loss/fused_linear_preference.py

@@ -32,6 +32,7 @@ class LigerFusedLinearPreferenceBase(torch.autograd.Function):
         alpha=1.0,
         beta=0.1,
         compiled=True,
+        **loss_kwargs,
     ):
         """
         Base class for fused linear layer with preference loss.
@@ -49,6 +50,7 @@ class LigerFusedLinearPreferenceBase(torch.autograd.Function):
             alpha (float): Weight for the NLL loss.
             beta (float): Weight for the odds ratio loss.
             compiled (bool): Whether to use torch compile for chunk accumulation.
+            loss_kwargs (dict): Other possible arguments that a loss function might need
         """
         # TODO: Tune CHUNK_SIZE to fully utilize the GPU
         CHUNK_SIZE = chunk_size
@@ -68,6 +70,7 @@ class LigerFusedLinearPreferenceBase(torch.autograd.Function):
             beta=beta,
             compute_nll_loss=compute_nll_loss,
             full_target=target,
+            **loss_kwargs,
         )
 
         def accumulate_chunk(input_chunk, target_chunk):
@@ -94,6 +97,9 @@ class LigerFusedLinearPreferenceBase(torch.autograd.Function):
             loss_acc.add_(chunk_loss)
             return chunk_grad_input
 
+        if compiled:
+            accumulate_chunk = torch.compile(accumulate_chunk)
+
         len_chosen = target.shape[0] // 2
         _chosen_input_chunks = torch.chunk(_input[:len_chosen], chunks=chunks, dim=0)
         _chosen_target_chunks = torch.chunk(target[:len_chosen], chunks=chunks, dim=0)
@@ -116,8 +122,6 @@ class LigerFusedLinearPreferenceBase(torch.autograd.Function):
                 [chosen_target_chunk, rejected_target_chunk], dim=0
             )
 
-            if compiled:
-                accumulate_chunk = torch.compile(accumulate_chunk)
             grad_input = accumulate_chunk(input_chunk, target_chunk)
 
             grad_chosen_inputs.append(grad_input[: chosen_target_chunk.shape[0]])

liger_kernel/chunked_loss/orpo_loss.py

@@ -34,7 +34,7 @@ class LigerFusedLinearORPOFunction(LigerFusedLinearPreferenceBase):
         ignore_index=-100,
         beta=0.1,
         compute_nll_loss=True,
-        compiled=True,
+        compiled=False,
     ):
         """
         Fused linear layer with ORPO (Odds-Ratio Preference Optimization) loss.

liger_kernel/chunked_loss/simpo_loss.py

@@ -0,0 +1,64 @@
+import torch.nn.functional as F
+
+from liger_kernel.chunked_loss.fused_linear_preference import (
+    LigerFusedLinearPreferenceBase,
+)
+
+
+class LigerFusedLinearSimPOFunction(LigerFusedLinearPreferenceBase):
+
+    @staticmethod
+    def preference_loss_fn(chosen_logps, rejected_logps, beta=0.1, gamma=0.5):
+        """
+        Compute odds-ratio loss.
+        Args:
+            chosen_logps (torch.Tensor): Avg log probabilities of chosen tokens. Shape: (batch_size,).
+            rejected_logps (torch.Tensor): Avg log probabilities of rejected tokens. Shape: (batch_size,).
+            beta (float): Weight for the odds ratio loss.
+            gamma (float): The simpo gamma, margin term.
+        """
+        logits = beta * (chosen_logps - rejected_logps) - gamma
+        loss = F.logsigmoid(logits).mean()
+        return loss
+
+    @staticmethod
+    def forward(
+        ctx,
+        _input,
+        weight,
+        target,
+        bias=None,
+        ignore_index=-100,
+        beta=0.1,
+        alpha=1.0,
+        compute_nll_loss=False,
+        compiled=True,
+        gamma=0.5,
+    ):
+        """
+        Fused linear layer with SimPO (Simple Preference Optimization) loss. https://arxiv.org/pdf/2405.14734
+        Handles both the forward and backward pass of the final linear layer with SimPO loss.
+        Inspired from LigerFusedLinearCrossEntropyFunction (https://arxiv.org/abs/2410.10989) which fuses final linear layer and CE loss.
+        """
+
+        return LigerFusedLinearPreferenceBase.forward(
+            ctx,
+            _input,
+            weight,
+            target,
+            bias,
+            loss_fn=LigerFusedLinearSimPOFunction.preference_loss_fn,
+            compute_nll_loss=compute_nll_loss,
+            ignore_index=ignore_index,
+            alpha=alpha,
+            beta=beta,
+            compiled=compiled,
+            gamma=gamma,
+        )
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        # Get gradients for _input, weight, bias, and target from the base class
+        grads = LigerFusedLinearPreferenceBase.backward(ctx, grad_output)[:4]
+        # Return these gradients, followed by None for the remaining inputs
+        return *grads, None, None, None, None, None, None

{liger_kernel_nightly-0.4.2.dev20241119054729.dist-info → liger_kernel_nightly-0.4.2.dev20241119174706.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.4.2.dev20241119054729
+Version: 0.4.2.dev20241119174706
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 

{liger_kernel_nightly-0.4.2.dev20241119054729.dist-info → liger_kernel_nightly-0.4.2.dev20241119174706.dist-info}/RECORD

@@ -2,8 +2,9 @@ liger_kernel/env_report.py,sha256=jye8RvUkmhqaIshdeIpoUABoAu7FPKJUib4FnAfvkpw,11
 liger_kernel/chunked_loss/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 liger_kernel/chunked_loss/cpo_loss.py,sha256=ty3nAlpxGqH6HMvTDzNOVulwvs-j6k26FIgEK0nl9Rc,2059
 liger_kernel/chunked_loss/dpo_loss.py,sha256=_sftycUsxypLiQaCIoqMEwtc425Kxiq97YI6DvFvscc,1943
-liger_kernel/chunked_loss/fused_linear_preference.py,sha256=9bj4u328TeTBGr5DquYU5sHANagXa3ti-6rjTIa-OXQ,8595
-liger_kernel/chunked_loss/orpo_loss.py,sha256=DNifPpzGV_t3dfOPlPy2XKDM6M1Qne0kCbIPztvFY9U,2179
+liger_kernel/chunked_loss/fused_linear_preference.py,sha256=gtsWG3rpTlWpiiom_oMPeS-w-lofBVrguN0KglAXTGk,8727
+liger_kernel/chunked_loss/orpo_loss.py,sha256=QtHPDQwZdU7QFgu9tPg81vQfF_Dm3zQcsmhp9SdKKvA,2180
+liger_kernel/chunked_loss/simpo_loss.py,sha256=lmPopkHcqfglEnXv28FcQQjIkpNg8CEhn0Wt19xcoE4,2223
 liger_kernel/ops/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 liger_kernel/ops/cross_entropy.py,sha256=sfUb7-jIZp0EKXjg1DYy2Wdzw_Mg-mHmGoR5bpdm4tw,15526
 liger_kernel/ops/fused_linear_cross_entropy.py,sha256=ib7M3AjJE164yMfuS9R39k-5qnDgYOXptIT146lqYbg,9964
@@ -50,9 +51,9 @@ liger_kernel/transformers/model/qwen2.py,sha256=EyhSSzQOskGjSnCsKMZpd1s5IAIlHd5P
 liger_kernel/transformers/model/qwen2_vl.py,sha256=bIQe2bWiY--G84FhCD29Gdi64_qHP6vbcGsK6vKysQE,8547
 liger_kernel/triton/__init__.py,sha256=yfRe0zMb47QnqjecZWG7LnanfCTzeku7SgWRAwNVmzU,101
 liger_kernel/triton/monkey_patch.py,sha256=5BcGKTtdqeYchypBIBopGIWPx1-cFALz7sOKoEsqXJ0,1584
-liger_kernel_nightly-0.4.2.dev20241119054729.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
-liger_kernel_nightly-0.4.2.dev20241119054729.dist-info/METADATA,sha256=z7a8F_CpxQybJzgLVPeWKTiLuv4nWL7gvQe4vIVzn7Q,21556
-liger_kernel_nightly-0.4.2.dev20241119054729.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
-liger_kernel_nightly-0.4.2.dev20241119054729.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-liger_kernel_nightly-0.4.2.dev20241119054729.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
-liger_kernel_nightly-0.4.2.dev20241119054729.dist-info/RECORD,,
+liger_kernel_nightly-0.4.2.dev20241119174706.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
+liger_kernel_nightly-0.4.2.dev20241119174706.dist-info/METADATA,sha256=T8u7FqqGYIi-w-4gDqYqgRq5y5vgV7Nk9FdcHYNlQxQ,21556
+liger_kernel_nightly-0.4.2.dev20241119174706.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
+liger_kernel_nightly-0.4.2.dev20241119174706.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+liger_kernel_nightly-0.4.2.dev20241119174706.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
+liger_kernel_nightly-0.4.2.dev20241119174706.dist-info/RECORD,,