liger-kernel-nightly 0.5.3.dev20250219232423__py3-none-any.whl → 0.5.3.dev20250220195514__py3-none-any.whl

liger_kernel/ops/cross_entropy.py

@@ -289,9 +289,9 @@ def cross_entropy_forward(
     weight_sum = 0.0
     if weight is not None:
         assert weight.shape[0] == V, f"If given, weight has to be a Tensor of size V. Got: {weight.shape}"
-        assert torch.is_floating_point(
-            weight
-        ), f"If given, weight has to be a Tensor of floating point dtype. Got: {weight.dtype}"
+        assert torch.is_floating_point(weight), (
+            f"If given, weight has to be a Tensor of floating point dtype. Got: {weight.dtype}"
+        )
         sum_non_ignore_weight = torch.gather(weight, dim=0, index=target.masked_select(target_mask)).sum().item()
         weight_sum = weight.sum().item()
         # ensure weight is contiguous

liger_kernel/ops/fused_linear_cross_entropy.py

@@ -58,9 +58,9 @@ def fused_linear_cross_entropy_forward(
     ce_weight_sum = 0.0
     if ce_weight is not None:
         assert ce_weight.shape[0] == V, f"If given, weight has to be a Tensor of size V. Got: {ce_weight.shape}"
-        assert torch.is_floating_point(
-            ce_weight
-        ), f"If given, weight has to be a Tensor of floating point dtype. Got: {ce_weight.dtype}"
+        assert torch.is_floating_point(ce_weight), (
+            f"If given, weight has to be a Tensor of floating point dtype. Got: {ce_weight.dtype}"
+        )
         total_sum_non_ignore_ce_weight = (
             torch.gather(ce_weight, dim=0, index=target.masked_select(target_mask)).sum().item()
         )

liger_kernel/ops/fused_linear_jsd.py

@@ -195,9 +195,9 @@ class LigerFusedLinearJSDFunction(torch.autograd.Function):
         """
         has_label = False
         if shift_labels is not None:
-            assert shift_labels.shape == (
-                teacher_input.shape[0],
-            ), f"the shape of shift_labels must be (BT,). Got: {shift_labels.shape}"
+            assert shift_labels.shape == (teacher_input.shape[0],), (
+                f"the shape of shift_labels must be (BT,). Got: {shift_labels.shape}"
+            )
             shift_labels = shift_labels.contiguous()
             has_label = True
 

liger_kernel/ops/jsd.py

@@ -157,9 +157,9 @@ class LigerJSDFunction(torch.autograd.Function):
         """
         has_label = False
         if shift_labels is not None:
-            assert shift_labels.shape == (
-                _input.shape[0],
-            ), f"the shape of shift_labels must be (BT,). Got: {shift_labels.shape}"
+            assert shift_labels.shape == (_input.shape[0],), (
+                f"the shape of shift_labels must be (BT,). Got: {shift_labels.shape}"
+            )
             shift_labels = shift_labels.contiguous()
             has_label = True
 

liger_kernel/ops/layer_norm.py

@@ -147,9 +147,9 @@ def layer_norm_forward(X, W, B, eps):
     Y = torch.empty((n_rows, n_cols), dtype=X.dtype, device=X.device)
     Mean = torch.empty(n_rows, dtype=X.dtype, device=X.device)
     RSTD = torch.empty(n_rows, dtype=X.dtype, device=X.device)
-    assert (
-        X.shape[1] == W.shape[0]
-    ), f"Incompatible hidden size dimension between input tensor with shape[1] = {X.shape[1]} and weight tensor with shape[0] = {W.shape[0]}"
+    assert X.shape[1] == W.shape[0], (
+        f"Incompatible hidden size dimension between input tensor with shape[1] = {X.shape[1]} and weight tensor with shape[0] = {W.shape[0]}"
+    )
 
     _layer_norm_forward_kernel[(n_rows,)](
         Y,

liger_kernel/ops/utils.py

@@ -49,8 +49,7 @@ def calculate_settings(n):
     BLOCK_SIZE = triton.next_power_of_2(n)
     if BLOCK_SIZE > MAX_FUSED_SIZE:
         raise RuntimeError(
-            f"Cannot launch Triton kernel since n = {n} exceeds "
-            f"the recommended Triton blocksize = {MAX_FUSED_SIZE}."
+            f"Cannot launch Triton kernel since n = {n} exceeds the recommended Triton blocksize = {MAX_FUSED_SIZE}."
         )
 
     num_warps = 4

liger_kernel/transformers/cross_entropy.py

@@ -17,9 +17,9 @@ class LigerCrossEntropyLoss(torch.nn.Module):
         return_z_loss: bool = False,
     ):
         super().__init__()
-        assert (label_smoothing >= 0) and (
-            label_smoothing <= 1
-        ), f"label_smoothing must be between 0.0 and 1.0. Got: {label_smoothing}"
+        assert (label_smoothing >= 0) and (label_smoothing <= 1), (
+            f"label_smoothing must be between 0.0 and 1.0. Got: {label_smoothing}"
+        )
         assert reduction in {
             "mean",
             "sum",

liger_kernel/transformers/fused_linear_cross_entropy.py

@@ -17,9 +17,9 @@ class LigerFusedLinearCrossEntropyLoss(torch.nn.Module):
         return_z_loss: bool = False,
     ):
         super().__init__()
-        assert (label_smoothing >= 0) and (
-            label_smoothing <= 1
-        ), f"label_smoothing must be between 0.0 and 1.0. Got: {label_smoothing}"
+        assert (label_smoothing >= 0) and (label_smoothing <= 1), (
+            f"label_smoothing must be between 0.0 and 1.0. Got: {label_smoothing}"
+        )
         assert reduction in {
             "mean",
             "sum",

liger_kernel/transformers/group_norm.py

@@ -21,9 +21,9 @@ class LigerGroupNorm(nn.Module):
             "zeros",
         ], f"init_fn must be either 'ones' or 'zeros', got {init_fn}"
 
-        assert (
-            num_channels % num_groups == 0
-        ), f"Number of channels {num_channels} must be divisible by num_groups {num_groups}"
+        assert num_channels % num_groups == 0, (
+            f"Number of channels {num_channels} must be divisible by num_groups {num_groups}"
+        )
         self.num_channels = num_channels
         self.num_groups = num_groups
         self.eps = eps
@@ -34,9 +34,9 @@ class LigerGroupNorm(nn.Module):
     def forward(self, hidden_states):
         # hidden_states: (batch_size, num_channels, *)
         assert hidden_states.dim() >= 3, f"Input must have atleast 3 dimensions, got {hidden_states.dim()}"
-        assert (
-            hidden_states.size(1) == self.num_channels
-        ), f"Input tensor must have {self.num_channels} channels, got {hidden_states.size(1)}"
+        assert hidden_states.size(1) == self.num_channels, (
+            f"Input tensor must have {self.num_channels} channels, got {hidden_states.size(1)}"
+        )
         return LigerGroupNormFunction.apply(
             hidden_states,
             self.weight,

liger_kernel/transformers/monkey_patch.py

@@ -85,9 +85,9 @@ def apply_liger_kernel_to_llama(
         loaded. Default is None.
     """
 
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.llama import modeling_llama
     from transformers.models.llama.modeling_llama import LlamaModel
@@ -159,9 +159,9 @@ def apply_liger_kernel_to_mllama(
         loaded. Default is None.
     """
 
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.mllama import modeling_mllama
     from transformers.models.mllama.modeling_mllama import MllamaForCausalLM
@@ -261,9 +261,9 @@ def apply_liger_kernel_to_mistral(
         model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
         loaded. Default is None.
     """
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.mistral import modeling_mistral
     from transformers.models.mistral.modeling_mistral import MistralModel
@@ -321,9 +321,9 @@ def apply_liger_kernel_to_mixtral(
         loaded. Default is None.
     """
 
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.mixtral import modeling_mixtral
     from transformers.models.mixtral.modeling_mixtral import MixtralModel
@@ -393,9 +393,9 @@ def apply_liger_kernel_to_gemma(
         model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
         loaded. Default is None.
     """
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.gemma import modeling_gemma
     from transformers.models.gemma.modeling_gemma import GemmaModel
@@ -467,9 +467,9 @@ def apply_liger_kernel_to_gemma2(
         model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
         loaded. Default is None.
     """
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.gemma2 import modeling_gemma2
     from transformers.models.gemma2.modeling_gemma2 import Gemma2Model
@@ -544,9 +544,9 @@ def apply_liger_kernel_to_qwen2(
         model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
         loaded. Default is None.
     """
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.qwen2 import modeling_qwen2
     from transformers.models.qwen2.modeling_qwen2 import Qwen2Model
@@ -619,9 +619,9 @@ def apply_liger_kernel_to_qwen2_vl(
         model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
         loaded. Default is None.
     """
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.qwen2_vl import modeling_qwen2_vl
     from transformers.models.qwen2_vl.modeling_qwen2_vl import Qwen2VLModel
@@ -689,9 +689,9 @@ def apply_liger_kernel_to_phi3(
         model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
         loaded. Default is None.
     """
-    assert not (
-        cross_entropy and fused_linear_cross_entropy
-    ), "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
 
     from transformers.models.phi3 import modeling_phi3
     from transformers.models.phi3.modeling_phi3 import Phi3Model

{liger_kernel_nightly-0.5.3.dev20250219232423.dist-info → liger_kernel_nightly-0.5.3.dev20250220195514.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.5.3.dev20250219232423
+Version: 0.5.3.dev20250220195514
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 
@@ -191,6 +191,11 @@ y = orpo_loss(lm_head.weight, x, target)
 - `torch >= 2.5.0` Install according to the instruction in Pytorch official webpage.
 - `triton >= 3.0.0` Install from pypi. (e.g. `pip install triton==3.0.0`)
 
+```bash
+# Need to pass the url when installing
+pip install -e .[dev] --extra-index-url https://download.pytorch.org/whl/nightly/rocm6.2
+```
+
 ### Optional Dependencies
 
 - `transformers >= 4.x`: Required if you plan to use the transformers models patching APIs. The specific model you are working will dictate the minimum version of transformers.

{liger_kernel_nightly-0.5.3.dev20250219232423.dist-info → liger_kernel_nightly-0.5.3.dev20250220195514.dist-info}/RECORD

@@ -16,33 +16,33 @@ liger_kernel/chunked_loss/kto_loss.py,sha256=eVNW6HVCAm32shpfhbRlk92Flnjd7G32v0g
 liger_kernel/chunked_loss/orpo_loss.py,sha256=yjcrrbVeemLYodoSKT-FMSnaPtyKAZ3aOrvPD6tTY6Y,3617
 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=SRzAF9Ek84pBVFy3wqQZs7AhRoorKRIgQ-Td_rtl1Kk,18606
-liger_kernel/ops/fused_linear_cross_entropy.py,sha256=jm0bJLRp7UpE4MCa3yhvaSErP792Ze8nNgtUIbOw5sw,10910
-liger_kernel/ops/fused_linear_jsd.py,sha256=eKqaADj7LgWfoYqyH03tjrmhNTfJOF1Dhx_bWzBTnTU,9600
+liger_kernel/ops/cross_entropy.py,sha256=D6vFFloiuxFXoWfjlIjmfO3tVaWOiYmztw9FKAi5vdU,18608
+liger_kernel/ops/fused_linear_cross_entropy.py,sha256=1Y3Uk_TCSjqKgoG2eot1ptnWXJXXQESqGvOmqAW1gsM,10912
+liger_kernel/ops/fused_linear_jsd.py,sha256=Seshez2qaM6HiTQ8_HEqSwhaeVruNT1SvIM4ZrAPBEU,9602
 liger_kernel/ops/geglu.py,sha256=axGvCIvlBzuluoAIrWTsp2iZM4BFKNInkPov8YVvH9E,4126
 liger_kernel/ops/group_norm.py,sha256=qD4D4lSjSgVtO52EBNLC2iTseALRgPgqXE50U2woggk,10837
-liger_kernel/ops/jsd.py,sha256=WwGY9ozuH3PMg3udRI6H96UqAEzIozJoO2HtHg7010M,6107
+liger_kernel/ops/jsd.py,sha256=0jNeRxpcNI5ckxCdoCNyO5GEedLIuzx3lz6KAiksc4o,6109
 liger_kernel/ops/kl_div.py,sha256=MnfuYqqQESON1X2Swy064x1urKtMFdgeSWd60VttBXI,8420
-liger_kernel/ops/layer_norm.py,sha256=quvt2zcwcJCDxrgm-iWoHzDYOoeZdMC76nZ_ckw6-p8,7640
+liger_kernel/ops/layer_norm.py,sha256=o5X_N0XNX0t-1AV3dyv43G0KJSyclUxcpNXzHNh35ks,7640
 liger_kernel/ops/qwen2vl_mrope.py,sha256=3GExhYpLgB4VUtyZyjRk8XjEur3W4EWF6HQ67ML5vBU,8481
 liger_kernel/ops/rms_norm.py,sha256=PWLJcdIKU5e-8BuYFHd9Cqlq6wmr6fUXKi9zQD4LetU,11727
 liger_kernel/ops/rope.py,sha256=ofmBOkUpZZO-Q8Z5B_LOFYYLD-YT-8WnJ4vGOrDYouI,8943
 liger_kernel/ops/swiglu.py,sha256=KmgMjaJQnbLLgZn2nEpbwHU_xpnYRweCyrLQSVvM1vA,3015
-liger_kernel/ops/utils.py,sha256=vMWxfcw02xUvjpEXQQ3Rrj68ddZ8Of3hiOmEFq1zSKg,3852
+liger_kernel/ops/utils.py,sha256=uoFKQqo-34N2TWQNvXMFywqGiOMMXNEVBxVojzlUAa0,3836
 liger_kernel/ops/experimental/embedding.py,sha256=tolj3tItkzpSb30zWqDN2_yX4ectflaQ8HMyKyFIQc8,4172
 liger_kernel/ops/experimental/mm_int8int2.py,sha256=TrS9lpwekrik_w5qE7AhMJD1bcq-OidjtbsW80oZ6IM,13314
 liger_kernel/transformers/__init__.py,sha256=QPmYkL6hosBPpPqCUGqvIvAtD9XzLgvZqZxUyYMZeVk,2008
 liger_kernel/transformers/auto_model.py,sha256=0qCTRZt280Bj_LcFdzo9hlaR-BWNazawXOGgoCZjgEg,1545
-liger_kernel/transformers/cross_entropy.py,sha256=LtiHlj_tK2YFpilwvbG_NEVzbf82zKRpWCZMjaFUd4M,1681
+liger_kernel/transformers/cross_entropy.py,sha256=z3KTWQnFxr_IZaVjtYt0ZNEWQdDdYThN35xWkHlDGH0,1683
 liger_kernel/transformers/functional.py,sha256=lDOjch622dJIc78K3ePFK_H1DX00GC5kKjodjcbEgbM,4624
-liger_kernel/transformers/fused_linear_cross_entropy.py,sha256=ygU7cycCHWvSrrTgn2TseN8t-Qwfer4V7ldwhZ1E6_g,1776
+liger_kernel/transformers/fused_linear_cross_entropy.py,sha256=09Rt7FZzLH42VOcIbQ4dlQd0o3Rlb4vk6fqiOQ7WTD8,1778
 liger_kernel/transformers/fused_linear_jsd.py,sha256=bZ4otCvWBuOnA5XdQL-FzZVItJlDt-ht9e_pG7PG93E,3999
 liger_kernel/transformers/geglu.py,sha256=mrgqzIUVd6lN7fkDKLkw5YaESDxDtFgbot430WwPVOQ,1107
-liger_kernel/transformers/group_norm.py,sha256=URmjkQFsrbMffzcJiGpX7ckxWlpL95AiJS-80hwAWPk,2173
+liger_kernel/transformers/group_norm.py,sha256=6qMAWOprr4SzP0YhNVNGQIBpM5aUHplUD2VuGJrMBz0,2173
 liger_kernel/transformers/jsd.py,sha256=DGqRnxIZxsvxo0_tbbxX3b-sDbDjC_yKufyRIHCcScY,2979
 liger_kernel/transformers/kl_div.py,sha256=WLffFbh1EExD2Eb1F7lN11fo9JJC-0751WJjZAF1Fj8,409
 liger_kernel/transformers/layer_norm.py,sha256=c9pk3PEasOKYR0rhe5e5nNrnYKVCEW4VC8S6LpCq9EQ,906
-liger_kernel/transformers/monkey_patch.py,sha256=6eXmtERKr4YUppRAaH7a_ml3AOz0ao68E8QnOyXtIkY,37794
+liger_kernel/transformers/monkey_patch.py,sha256=DXU00zsQvSjAqCx7l36gKm1O81FuHgILkZMhyx4ZSys,37812
 liger_kernel/transformers/qwen2vl_mrope.py,sha256=5EwSqrMdsL9MYspeBMXBsNJKvH0MOmRrtJXAJlnnlOI,1047
 liger_kernel/transformers/rms_norm.py,sha256=GqCEJuGt0YdqqlMcToE0Wp4A8YFquDa4UUSyH2uFW2A,1191
 liger_kernel/transformers/rope.py,sha256=ZTrTORSAyfcFIKjk6XEeYmk4ROH7xXED9L4g2NFntlE,999
@@ -63,9 +63,9 @@ liger_kernel/transformers/trainer/__init__.py,sha256=p7yQfklV8-467qSz_ZMimkbDF7H
 liger_kernel/transformers/trainer/orpo_trainer.py,sha256=pdekW7l6Qg_aqa5SYKYlSWUF8m3lkOFvFLcIMEHrz9s,8338
 liger_kernel/triton/__init__.py,sha256=qCiCamzCRv6lpV8IqpAc9YMdNKC7GKurClWceQPnlis,92
 liger_kernel/triton/monkey_patch.py,sha256=Rd0hUHAzDkFfHvnX7-PBaNK5EKnZhtfM_h-fgQH9HPY,1568
-liger_kernel_nightly-0.5.3.dev20250219232423.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
-liger_kernel_nightly-0.5.3.dev20250219232423.dist-info/METADATA,sha256=cwUbT2K8osL4Xudlk83LjnQ3vtiZ1Ip8qhvSrXYcaTo,21625
-liger_kernel_nightly-0.5.3.dev20250219232423.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
-liger_kernel_nightly-0.5.3.dev20250219232423.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-liger_kernel_nightly-0.5.3.dev20250219232423.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
-liger_kernel_nightly-0.5.3.dev20250219232423.dist-info/RECORD,,
+liger_kernel_nightly-0.5.3.dev20250220195514.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
+liger_kernel_nightly-0.5.3.dev20250220195514.dist-info/METADATA,sha256=WkDgh3E1y7TYWCDttILZqikHz2S5b2kxLKoJ7JiWMd8,21766
+liger_kernel_nightly-0.5.3.dev20250220195514.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
+liger_kernel_nightly-0.5.3.dev20250220195514.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+liger_kernel_nightly-0.5.3.dev20250220195514.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
+liger_kernel_nightly-0.5.3.dev20250220195514.dist-info/RECORD,,