liger-kernel-nightly 0.5.2.dev20250124002122__tar.gz → 0.5.2.dev20250127064410__tar.gz

{liger_kernel_nightly-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.5.2.dev20250124002122
+Version: 0.5.2.dev20250127064410
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 

{liger_kernel_nightly-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "liger_kernel_nightly"
-version = "0.5.2.dev20250124002122"
+version = "0.5.2.dev20250127064410"
 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-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/src/liger_kernel_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.5.2.dev20250124002122
+Version: 0.5.2.dev20250127064410
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 

{liger_kernel_nightly-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/test/convergence/test_mini_models_multimodal.py

@@ -346,10 +346,13 @@ def run_mini_model_multimodal(
             1e-5,
             5e-3,
             1e-5,
-            marks=pytest.mark.skipif(
-                not QWEN2_VL_AVAILABLE,
-                reason="Qwen2-VL not available in this version of transformers",
-            ),
+            marks=[
+                pytest.mark.skipif(
+                    not QWEN2_VL_AVAILABLE,
+                    reason="Qwen2-VL not available in this version of transformers",
+                ),
+                pytest.mark.skipif(device == "xpu", reason="skip for XPU"),
+            ],
         ),
         pytest.param(
             "mini_qwen2_vl",
@@ -368,6 +371,7 @@ def run_mini_model_multimodal(
                     not QWEN2_VL_AVAILABLE,
                     reason="Qwen2-VL not available in this version of transformers",
                 ),
+                pytest.mark.skipif(device == "xpu", reason="skip for XPU"),
             ],
         ),
         pytest.param(

{liger_kernel_nightly-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/test/transformers/test_fused_linear_cross_entropy.py

@@ -105,7 +105,7 @@ class LigerLMHeadCE(torch.nn.Module):
 @pytest.mark.parametrize(
     "B, T, H, V",
     [
-        (8, 128, 1024, 4096),
+        pytest.param(8, 128, 1024, 4096, marks=pytest.mark.skipif(device="xpu", reason="skip for XPU")),
         (4, 47, 31, 123),  # random shape
     ],
 )
@@ -287,7 +287,7 @@ def test_correctness_functional(B, T, H, V, scalar, dtype, bias, ce_weight, atol
 @pytest.mark.parametrize(
     "B, T, H, V",
     [
-        (8, 128, 1024, 4096),
+        pytest.param(8, 128, 1024, 4096, marks=pytest.mark.skipif(device="xpu", reason="skip for XPU")),
         (4, 47, 31, 123),  # random shape
     ],
 )

{liger_kernel_nightly-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/test/transformers/test_rms_norm.py

@@ -103,7 +103,7 @@ class GemmaRMSNorm(nn.Module):
     [
         (LlamaRMSNorm, 0.0, "llama"),
         (GemmaRMSNorm, 1.0, "gemma"),
-        (BaseRMSNorm, 0.0, "none"),
+        pytest.param(BaseRMSNorm, 0.0, "none", marks=pytest.mark.skipif(device="xpu", reason="skip for XPU")),
     ],
 )
 @pytest.mark.parametrize(

{liger_kernel_nightly-0.5.2.dev20250124002122 → liger_kernel_nightly-0.5.2.dev20250127064410}/test/utils.py

@@ -541,7 +541,9 @@ class HFAlignmentLoss:
         **loss_kwargs,
     ):
         """Compute the loss metrics for the given batch of inputs for train or test."""
-        forward_output = self.concatenated_forward(_input, weight, target, bias, average_log_prob, preference_labels, nll_target)
+        forward_output = self.concatenated_forward(
+            _input, weight, target, bias, average_log_prob, preference_labels, nll_target
+        )
         (
             policy_chosen_logps,
             policy_rejected_logps,