liger-kernel-nightly 0.6.2.dev20250913213534__tar.gz → 0.6.2.dev20250919191028__tar.gz

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.6.2.dev20250913213534
+Version: 0.6.2.dev20250919191028
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 
@@ -35,9 +35,7 @@ Requires-Dist: triton>=2.3.1
 Provides-Extra: dev
 Requires-Dist: transformers>=4.49.0; extra == "dev"
 Requires-Dist: matplotlib>=3.7.2; extra == "dev"
-Requires-Dist: flake8>=4.0.1.1; extra == "dev"
-Requires-Dist: black>=24.4.2; extra == "dev"
-Requires-Dist: isort>=5.13.2; extra == "dev"
+Requires-Dist: ruff>=0.12.0; extra == "dev"
 Requires-Dist: pytest>=7.1.2; extra == "dev"
 Requires-Dist: pytest-xdist; extra == "dev"
 Requires-Dist: pytest-cov; extra == "dev"
@@ -45,7 +43,6 @@ Requires-Dist: pytest-asyncio; extra == "dev"
 Requires-Dist: pytest-rerunfailures; extra == "dev"
 Requires-Dist: datasets>=2.19.2; extra == "dev"
 Requires-Dist: seaborn; extra == "dev"
-Requires-Dist: mkdocs; extra == "dev"
 Requires-Dist: mkdocs-material; extra == "dev"
 Requires-Dist: torchvision>=0.20; extra == "dev"
 

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/docs/Examples.md

@@ -239,7 +239,7 @@ from liger_kernel.transformers.trainer import LigerORPOTrainer  # noqa: F401
 
 model = AutoModelForCausalLM.from_pretrained(
     "meta-llama/Llama-3.2-1B-Instruct",
-    torch_dtype=torch.bfloat16,
+    dtype=torch.bfloat16,
 )
 
 tokenizer = AutoTokenizer.from_pretrained(

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/examples/alignment/run_orpo.py

@@ -9,7 +9,7 @@ from liger_kernel.transformers.trainer import LigerORPOTrainer  # noqa: F401
 
 model = AutoModelForCausalLM.from_pretrained(
     "meta-llama/Llama-3.2-1B-Instruct",
-    torch_dtype=torch.bfloat16,
+    dtype=torch.bfloat16,
 )
 
 tokenizer = AutoTokenizer.from_pretrained(

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/examples/huggingface/training.py

@@ -48,7 +48,7 @@ def train():
             custom_args.model_name,
             trust_remote_code=True,
             use_cache=False,
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             # These args will get passed to the appropriate apply_liger_kernel_to_* function
             # to override the default settings
             # cross_entropy=True,
@@ -59,7 +59,7 @@ def train():
             custom_args.model_name,
             trust_remote_code=True,
             use_cache=False,
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
         )
 
     trainer = SFTTrainer(

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/examples/huggingface/training_multimodal.py

@@ -56,7 +56,7 @@ def construct_model_and_processor(model_name: str, use_liger: bool) -> torch.nn.
         model = Qwen2VLForConditionalGeneration.from_pretrained(
             pretrained_model_name_or_path=model_name,
             use_cache=False,
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             low_cpu_mem_usage=True,
             attn_implementation="sdpa",
         )

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/examples/medusa/train.py

@@ -319,7 +319,7 @@ def train():
         model = model_builder(
             model_args.model_name_or_path,
             cache_dir=training_args.cache_dir,
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
         )
 
         # Freeze the base model

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "liger_kernel_nightly"
-version = "0.6.2.dev20250913213534"
+version = "0.6.2.dev20250919191028"
 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-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/setup.py

@@ -33,9 +33,7 @@ def get_optional_dependencies():
         "dev": [
             "transformers>=4.49.0",
             "matplotlib>=3.7.2",
-            "flake8>=4.0.1.1",
-            "black>=24.4.2",
-            "isort>=5.13.2",
+            "ruff>=0.12.0",
             "pytest>=7.1.2",
             "pytest-xdist",
             "pytest-cov",
@@ -43,7 +41,6 @@ def get_optional_dependencies():
             "pytest-rerunfailures",
             "datasets>=2.19.2",
             "seaborn",
-            "mkdocs",
             "mkdocs-material",
             "torchvision>=0.20",
         ]

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/src/liger_kernel/transformers/fused_linear_cross_entropy.py

@@ -25,7 +25,7 @@ class LigerFusedLinearCrossEntropyLoss(torch.nn.Module):
         assert reduction in {
             "mean",
             "sum",
-            "none", 
+            "none",
         }, f"reduction must be 'mean' or 'sum' or 'none'. Got: {reduction}"
         assert softcap is None or softcap > 0, f"softcap must greater than 0.0 or None. Got: {softcap}"
         self.ce_weight = ce_weight

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/src/liger_kernel/transformers/model/glm4v.py

@@ -70,7 +70,7 @@ def lce_forward(
     >>> processor = AutoProcessor.from_pretrained(MODEL_PATH, use_fast=True)
     >>> model = Glm4vForConditionalGeneration.from_pretrained(
         pretrained_model_name_or_path=MODEL_PATH,
-        torch_dtype=torch.bfloat16,
+        dtype=torch.bfloat16,
         device_map="auto",
     )
     >>> inputs = processor.apply_chat_template(

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/src/liger_kernel/transformers/model/glm4v_moe.py

@@ -75,7 +75,7 @@ def lce_forward(
     >>> processor = AutoProcessor.from_pretrained(MODEL_PATH)
     >>> model = Glm4vMoeForConditionalGeneration.from_pretrained(
         pretrained_model_name_or_path=MODEL_PATH,
-        torch_dtype="auto",
+        dtype="auto",
         device_map="auto",
     )
     >>> inputs = processor.apply_chat_template(

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/src/liger_kernel_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.6.2.dev20250913213534
+Version: 0.6.2.dev20250919191028
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 
@@ -35,9 +35,7 @@ Requires-Dist: triton>=2.3.1
 Provides-Extra: dev
 Requires-Dist: transformers>=4.49.0; extra == "dev"
 Requires-Dist: matplotlib>=3.7.2; extra == "dev"
-Requires-Dist: flake8>=4.0.1.1; extra == "dev"
-Requires-Dist: black>=24.4.2; extra == "dev"
-Requires-Dist: isort>=5.13.2; extra == "dev"
+Requires-Dist: ruff>=0.12.0; extra == "dev"
 Requires-Dist: pytest>=7.1.2; extra == "dev"
 Requires-Dist: pytest-xdist; extra == "dev"
 Requires-Dist: pytest-cov; extra == "dev"
@@ -45,7 +43,6 @@ Requires-Dist: pytest-asyncio; extra == "dev"
 Requires-Dist: pytest-rerunfailures; extra == "dev"
 Requires-Dist: datasets>=2.19.2; extra == "dev"
 Requires-Dist: seaborn; extra == "dev"
-Requires-Dist: mkdocs; extra == "dev"
 Requires-Dist: mkdocs-material; extra == "dev"
 Requires-Dist: torchvision>=0.20; extra == "dev"
 

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/src/liger_kernel_nightly.egg-info/requires.txt

@@ -4,9 +4,7 @@ triton>=2.3.1
 [dev]
 transformers>=4.49.0
 matplotlib>=3.7.2
-flake8>=4.0.1.1
-black>=24.4.2
-isort>=5.13.2
+ruff>=0.12.0
 pytest>=7.1.2
 pytest-xdist
 pytest-cov
@@ -14,6 +12,5 @@ pytest-asyncio
 pytest-rerunfailures
 datasets>=2.19.2
 seaborn
-mkdocs
 mkdocs-material
 torchvision>=0.20

{liger_kernel_nightly-0.6.2.dev20250913213534 → liger_kernel_nightly-0.6.2.dev20250919191028}/test/transformers/test_monkey_patch.py

@@ -338,7 +338,7 @@ def test_apply_liger_kernel_to_instance_for_llama():
     with patch("transformers.models.llama.modeling_llama"):
         # Instantiate a dummy model
         config = transformers.models.llama.configuration_llama.LlamaConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -382,7 +382,7 @@ def test_apply_liger_kernel_to_instance_for_mllama_for_conditional_generation():
 
         # Instantiate a dummy model
         config = transformers.models.mllama.configuration_mllama.MllamaConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             text_config=transformers.models.mllama.configuration_mllama.MllamaTextConfig(
                 rms_norm_eps=1e-5,
                 hidden_size=32,
@@ -533,7 +533,7 @@ def test_apply_liger_kernel_to_instance_for_llama4_for_causal_lm():
 
         # Instantiate a dummy model
         config = transformers.models.llama4.configuration_llama4.Llama4TextConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -573,9 +573,9 @@ def test_apply_liger_kernel_to_instance_for_llama4_for_conditional_generation():
 
         # Instantiate a dummy model
         config = transformers.models.llama4.configuration_llama4.Llama4Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             text_config=transformers.models.llama4.configuration_llama4.Llama4TextConfig(
-                torch_dtype=torch.bfloat16,
+                dtype=torch.bfloat16,
                 rms_norm_eps=1e-5,
                 hidden_size=32,
                 intermediate_size=64,
@@ -656,7 +656,7 @@ def test_apply_liger_kernel_to_instance_for_mistral():
     with patch("transformers.models.mistral.modeling_mistral"):
         # Instantiate a dummy model
         config = transformers.models.mistral.configuration_mistral.MistralConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -695,7 +695,7 @@ def test_apply_liger_kernel_to_instance_for_mixtral():
     with patch("transformers.models.mixtral.modeling_mixtral"):
         # Instantiate a dummy model
         config = transformers.models.mixtral.configuration_mixtral.MixtralConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -738,7 +738,7 @@ def test_apply_liger_kernel_to_instance_for_gemma():
     with patch("transformers.models.gemma.modeling_gemma"):
         # Instantiate a dummy model
         config = transformers.models.gemma.configuration_gemma.GemmaConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -777,7 +777,7 @@ def test_apply_liger_kernel_to_instance_for_gemma2():
     with patch("transformers.models.gemma2.modeling_gemma2"):
         # Instantiate a dummy model
         config = transformers.models.gemma2.configuration_gemma2.Gemma2Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -827,7 +827,7 @@ def test_apply_liger_kernel_to_instance_for_paligemma():
 
         # Instantiate a dummy model
         config = transformers.models.paligemma.configuration_paligemma.PaliGemmaConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             text_config={
                 "num_hidden_layers": 2,
                 "rms_norm_eps": 1e-5,
@@ -883,7 +883,7 @@ def test_apply_liger_kernel_to_instance_for_gemma3_text():
 
         # Instantiate a dummy model
         config = transformers.models.gemma3.configuration_gemma3.Gemma3TextConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -939,7 +939,7 @@ def test_apply_liger_kernel_to_instance_for_gemma3_conditional_generation():
 
         # Instantiate a dummy model
         text_config = transformers.models.gemma3.configuration_gemma3.Gemma3TextConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1026,7 +1026,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2():
     with patch("transformers.models.qwen2.modeling_qwen2"):
         # Instantiate a dummy model
         config = transformers.models.qwen2.configuration_qwen2.Qwen2Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1068,7 +1068,7 @@ def test_apply_liger_kernel_to_instance_for_qwen3():
 
         # Instantiate a dummy model
         config = transformers.models.qwen3.configuration_qwen3.Qwen3Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1110,7 +1110,7 @@ def test_apply_liger_kernel_to_instance_for_qwen3_moe():
 
         # Instantiate a dummy model
         config = transformers.models.qwen3_moe.configuration_qwen3_moe.Qwen3MoeConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1158,7 +1158,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2_vl_for_conditional_generation(
 
         # Instantiate a dummy model
         config = transformers.models.qwen2_vl.configuration_qwen2_vl.Qwen2VLConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=48,
@@ -1227,7 +1227,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2_vl():
 
         # Instantiate a dummy model
         config = transformers.models.qwen2_vl.configuration_qwen2_vl.Qwen2VLConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=48,
@@ -1294,7 +1294,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2_vl_text():
 
         # Instantiate a dummy model
         config = transformers.models.qwen2_vl.configuration_qwen2_vl.Qwen2VLTextConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=48,
@@ -1347,7 +1347,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2_5_vl():
 
         # Instantiate a dummy model
         config = transformers.models.qwen2_5_vl.configuration_qwen2_5_vl.Qwen2_5_VLConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=48,
@@ -1416,7 +1416,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2_5_vl_for_conditional_generatio
 
         # Instantiate a dummy model
         config = transformers.models.qwen2_5_vl.configuration_qwen2_5_vl.Qwen2_5_VLConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=48,
@@ -1483,7 +1483,7 @@ def test_apply_liger_kernel_to_instance_for_qwen2_5_vl_text():
 
         # Instantiate a dummy model
         config = transformers.models.qwen2_5_vl.configuration_qwen2_5_vl.Qwen2_5_VLTextConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=48,
@@ -1528,7 +1528,7 @@ def test_apply_liger_kernel_to_instance_for_phi3():
     with patch("transformers.models.phi3.modeling_phi3"):
         # Instantiate a dummy model
         config = transformers.models.phi3.configuration_phi3.Phi3Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1570,7 +1570,7 @@ def test_apply_liger_kernel_to_instance_for_olmo2():
 
         # Instantiate a dummy model
         config = transformers.models.olmo2.configuration_olmo2.Olmo2Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1616,7 +1616,7 @@ def test_apply_liger_kernel_to_instance_for_glm4():
 
         # Instantiate a dummy model
         config = transformers.models.glm4.configuration_glm4.Glm4Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,
@@ -1664,7 +1664,7 @@ def test_apply_liger_kernel_to_instance_for_glm4v():
 
         # Instantiate a dummy model
         config = transformers.models.glm4v.configuration_glm4v.Glm4vConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             text_config={
                 "num_hidden_layers": 2,
                 "rms_norm_eps": 1e-5,
@@ -1734,7 +1734,7 @@ def test_apply_liger_kernel_to_instance_for_glm4v_moe():
 
         # Instantiate a dummy model
         config = transformers.models.glm4v_moe.configuration_glm4v_moe.Glm4vMoeConfig(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             hidden_size=32,
             num_attention_heads=4,
             num_key_value_heads=2,
@@ -1837,7 +1837,7 @@ def test_apply_liger_kernel_to_instance_for_smollm3():
     with patch("transformers.models.smollm3.modeling_smollm3"):
         # Instantiate a dummy model
         config = transformers.models.smollm3.configuration_smollm3.SmolLM3Config(
-            torch_dtype=torch.bfloat16,
+            dtype=torch.bfloat16,
             rms_norm_eps=1e-5,
             hidden_size=32,
             intermediate_size=64,