liger-kernel 0.5.3__py3-none-any.whl → 0.5.4__py3-none-any.whl

liger_kernel/transformers/monkey_patch.py

@@ -61,6 +61,85 @@ def _patch_layer_norm_module(module, eps=1e-6):
     _bind_method_to_module(module, "extra_repr", LigerLayerNorm.extra_repr)
 
 
+def apply_liger_kernel_to_granite(
+    rope: bool = True,
+    cross_entropy: bool = True,
+    fused_linear_cross_entropy: bool = False,
+    rms_norm: bool = True,
+    swiglu: bool = True,
+    model: PreTrainedModel = None,
+) -> None:
+    """
+    Apply Liger kernels to replace original implementation in HuggingFace Granite 3 models
+
+    Args:
+        rope (bool): Whether to apply Liger's rotary position embedding. Default is True.
+        cross_entropy (bool): Whether to apply Liger's cross entropy loss. Default is True.
+        fused_linear_cross_entropy (bool):
+            Whether to apply Liger's fused linear cross entropy loss. Default is False.
+            `cross_entropy` and `fused_linear_cross_entropy` cannot both be True.
+            If `fused_linear_cross_entropy` is True, the logits will not be materialized but more memory efficient.
+        rms_norm (bool): Whether to apply Liger's RMSNorm. Default is True.
+        swiglu (bool): Whether to apply Liger's SwiGLU MLP. Default is True.
+        model (PreTrainedModel): The model instance to apply Liger kernels to, if the model has already been
+        loaded. Default is None.
+
+
+
+    Debugging notes:
+        If LigerSwiGLUMLP is OK for Llama, it should be fine for Granite, but it's not.
+    """
+
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
+
+    from transformers.models.granite import modeling_granite
+    from transformers.models.granite.modeling_granite import GraniteModel
+
+    if swiglu:
+        modeling_granite.GraniteMLP = LigerSwiGLUMLP
+
+    if rms_norm:
+        modeling_granite.GraniteRMSNorm = LigerRMSNorm
+
+    if rope:
+        modeling_granite.apply_rotary_pos_emb = liger_rotary_pos_emb
+
+    if cross_entropy:
+        if transformer_version >= version.parse(SUPPORTED_TRANSFORMER_VERSION):
+            from transformers.loss.loss_utils import nn
+
+            nn.functional.cross_entropy = liger_cross_entropy
+        else:
+            logger.warning(TRANSFORMER_DEPRECATION_WARNING)
+            modeling_granite.CrossEntropyLoss = LigerCrossEntropyLoss
+
+    if fused_linear_cross_entropy:
+        raise NotImplementedError("LigerFusedLinearCrossEntropy is not available for Granite models.")
+        # NOTE: Granite model `GraniteForCausalLM.forward` scales logits each
+        # call, so we can't sidestep logit materialization. A bit more work
+        # would be needed to add a scaling term to the `LigerFusedLinearCrossEntropyFunction`
+        # for the logit output.
+
+    if model is not None:
+        # The model instance already exists, so we need to additionally patch the
+        # instance variables that reference already-instantiated modules (e.g. GraniteRMSNorm or GraniteMLP)
+
+        # get the base model from the model instance
+        base_model: GraniteModel = getattr(model, model.base_model_prefix, model)
+
+        if rms_norm:
+            _patch_rms_norm_module(base_model.norm)
+
+        for decoder_layer in base_model.layers:
+            if swiglu:
+                _bind_method_to_module(decoder_layer.mlp, "forward", LigerSwiGLUMLP.forward)
+            if rms_norm:
+                _patch_rms_norm_module(decoder_layer.input_layernorm)
+                _patch_rms_norm_module(decoder_layer.post_attention_layernorm)
+
+
 def apply_liger_kernel_to_llama(
     rope: bool = True,
     cross_entropy: bool = False,
@@ -85,9 +164,9 @@ 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 +238,9 @@ 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 +340,9 @@ 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 +400,9 @@ 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 +472,9 @@ 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 +546,9 @@ 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 +623,9 @@ 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 +698,9 @@ 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 +768,9 @@ 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
@@ -735,15 +814,80 @@ def apply_liger_kernel_to_phi3(
                 _patch_rms_norm_module(decoder_layer.post_attention_layernorm)
 
 
+def apply_liger_kernel_to_olmo2(
+    rope: bool = True,
+    cross_entropy: bool = False,
+    fused_linear_cross_entropy: bool = True,
+    rms_norm: bool = True,
+    swiglu: bool = True,
+    model: PreTrainedModel = None,
+) -> None:
+    """
+    Apply Liger kernels to replace original implementation in HuggingFace OLMO2 models.
+
+    Args:
+        rope (bool): Whether to apply Liger's rotary position embedding. Default is True.
+        cross_entropy (bool): Whether to apply Liger's cross entropy loss. Default is False.
+        fused_linear_cross_entropy (bool):
+            Whether to apply Liger's fused linear cross entropy loss. Default is True.
+            `cross_entropy` and `fused_linear_cross_entropy` cannot both be True.
+            If `fused_linear_cross_entropy` is True, the logits will not be materialized but more memory efficient.
+        rms_norm (bool): Whether to apply Liger's RMSNorm. Default is True.
+        swiglu (bool): Whether to apply Liger's SwiGLU Olmo2MLP. Default is True.
+        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."
+    )
+
+    from transformers.models.olmo2 import modeling_olmo2
+    from transformers.models.olmo2.modeling_olmo2 import Olmo2Model
+
+    from liger_kernel.transformers.model.olmo2 import lce_forward as olmo2_lce_forward
+
+    if rope:
+        modeling_olmo2.apply_rotary_pos_emb = liger_rotary_pos_emb
+    if rms_norm:
+        modeling_olmo2.Olmo2RMSNorm = partial(LigerRMSNorm, in_place=False)
+    if swiglu:
+        modeling_olmo2.Olmo2MLP = LigerSwiGLUMLP
+    if cross_entropy:
+        from transformers.loss.loss_utils import nn
+
+        nn.functional.cross_entropy = liger_cross_entropy
+    if fused_linear_cross_entropy:
+        modeling_olmo2.Olmo2ForCausalLM.forward = olmo2_lce_forward
+
+    if model is not None:
+        # The model instance already exists, so we need to additionally patch the
+        # instance variables that reference already-instantiated modules
+
+        # get the base model from the model instance
+        base_model: Olmo2Model = getattr(model, model.base_model_prefix, model)
+
+        if rms_norm:
+            _patch_rms_norm_module(base_model.norm)
+
+        for decoder_layer in base_model.layers:
+            if swiglu:
+                _bind_method_to_module(decoder_layer.mlp, "forward", LigerSwiGLUMLP.forward)
+            if rms_norm:
+                _patch_rms_norm_module(decoder_layer.post_attention_layernorm, in_place=False)
+                _patch_rms_norm_module(decoder_layer.post_feedforward_layernorm, in_place=False)
+
+
 # Model type corresponds to the keys defined in transformers/models/auto/modeling_auto.py
 MODEL_TYPE_TO_APPLY_LIGER_FN = {
     "gemma": apply_liger_kernel_to_gemma,
     "gemma2": apply_liger_kernel_to_gemma2,
     "llama": apply_liger_kernel_to_llama,
+    "granite": apply_liger_kernel_to_granite,
     "mllama": apply_liger_kernel_to_mllama,
     "mllama_text_model": apply_liger_kernel_to_mllama,
     "mistral": apply_liger_kernel_to_mistral,
     "mixtral": apply_liger_kernel_to_mixtral,
+    "olmo2": apply_liger_kernel_to_olmo2,
     "qwen2": apply_liger_kernel_to_qwen2,
     "qwen2_vl": apply_liger_kernel_to_qwen2_vl,
     "phi3": apply_liger_kernel_to_phi3,

liger_kernel/transformers/tvd.py

@@ -0,0 +1,13 @@
+import torch.nn as nn
+
+from liger_kernel.ops.tvd import LigerTVDLossFunction
+
+
+class LigerTVDLoss(nn.Module):
+    def __init__(self, reduction="batchmean", ignore_index: int = -100):
+        super(LigerTVDLoss, self).__init__()
+        self.reduction = reduction
+        self.ignore_index = ignore_index
+
+    def forward(self, p, q, shift_labels=None):
+        return LigerTVDLossFunction.apply(p, q, shift_labels, self.reduction, self.ignore_index)

liger_kernel/utils.py

@@ -9,5 +9,54 @@ def infer_device():
         return "cuda"
     elif torch.xpu.is_available():
         return "xpu"
+    elif torch.hip.is_available():
+        return "hip"
     else:
         return "cpu"
+
+
+def transformers_version_dispatch(
+    required_version: str,
+    before_fn,
+    after_fn,
+    before_args: tuple = (),
+    after_args: tuple = (),
+    before_kwargs: dict = None,
+    after_kwargs: dict = None,
+):
+    """
+    Dispatches to different functions based on package version comparison.
+
+    Args:
+        required_version: Version to compare against (e.g. "4.48.0")
+        before_fn: Function to call if package_version < required_version
+        after_fn: Function to call if package_version >= required_version
+        before_args: Positional arguments for before_fn
+        after_args: Positional arguments for after_fn
+        before_kwargs: Keyword arguments for before_fn
+        after_kwargs: Keyword arguments for after_fn
+
+    Returns:
+        Result from either before_fn or after_fn
+
+    Example:
+        >>> rotary_emb = transformers_version_dispatch(
+        ...     "4.48.0",
+        ...     LlamaRotaryEmbedding,
+        ...     LlamaRotaryEmbedding,
+        ...     before_args=(head_dim,),
+        ...     after_args=(LlamaConfig(head_dim=head_dim),),
+        ...     before_kwargs={'device': device},
+        ...     after_kwargs={'device': device}
+        ... )
+    """
+    from packaging import version
+    from transformers import __version__ as transformers_version
+
+    before_kwargs = before_kwargs or {}
+    after_kwargs = after_kwargs or {}
+
+    if version.parse(transformers_version) < version.parse(required_version):
+        return before_fn(*before_args, **before_kwargs)
+    else:
+        return after_fn(*after_args, **after_kwargs)

{liger_kernel-0.5.3.dist-info → liger_kernel-0.5.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: liger_kernel
-Version: 0.5.3
+Version: 0.5.4
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 
@@ -97,6 +97,11 @@ Dynamic: requires-dist
                     <img src="https://github.com/linkedin/Liger-Kernel/actions/workflows/amd-ci.yml/badge.svg?event=schedule" alt="Build">
                 </a>
             </div>
+            <div style="display: block;">
+                <a href="https://github.com/linkedin/Liger-Kernel/actions/workflows/amd-ci.yml">
+                    <img src="https://github.com/linkedin/Liger-Kernel/actions/workflows/intel-ci.yml/badge.svg?event=schedule" alt="Build">
+                </a>
+            </div>
         </td>
     </tr>
 </table>
@@ -123,7 +128,7 @@ Dynamic: requires-dist
 
 **Liger Kernel** is a collection of Triton kernels designed specifically for LLM training. It can effectively increase multi-GPU **training throughput by 20%** and reduces **memory usage by 60%**. We have implemented **Hugging Face Compatible** `RMSNorm`, `RoPE`, `SwiGLU`, `CrossEntropy`, `FusedLinearCrossEntropy`, and more to come. The kernel works out of the box with [Flash Attention](https://github.com/Dao-AILab/flash-attention), [PyTorch FSDP](https://pytorch.org/tutorials/intermediate/FSDP_tutorial.html), and [Microsoft DeepSpeed](https://github.com/microsoft/DeepSpeed). We welcome contributions from the community to gather the best kernels for LLM training.
 
-We've also added optimized Post-Training kernels that deliver **up to 80% memory savings** for alignment and distillation tasks. We support losses like DPO, CPO, ORPO, SimPO, JSD, and many more. Check out [how we optimize the memory](https://x.com/hsu_byron/status/1866577403918917655).
+We've also added optimized Post-Training kernels that deliver **up to 80% memory savings** for alignment and distillation tasks. We support losses like DPO, CPO, ORPO, SimPO, KTO, JSD, and many more. Check out [how we optimize the memory](https://x.com/hsu_byron/status/1866577403918917655).
 
 ## Supercharge Your Model with Liger Kernel
 
@@ -188,6 +193,11 @@ 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.
@@ -305,6 +315,8 @@ loss.backward()
 | Qwen2, Qwen2.5, & QwQ      | `liger_kernel.transformers.apply_liger_kernel_to_qwen2`    | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy        |
 | Qwen2-VL, & QVQ       | `liger_kernel.transformers.apply_liger_kernel_to_qwen2_vl`    | RMSNorm, LayerNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy        |
 | Phi3 & Phi3.5       | `liger_kernel.transformers.apply_liger_kernel_to_phi3`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy         |
+| Granite 3.0 & 3.1   | `liger_kernel.transformers.apply_liger_kernel_to_granite`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss |
+| OLMo2   | `liger_kernel.transformers.apply_liger_kernel_to_olmo2`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
 
 
 ## Low-level APIs
@@ -333,6 +345,7 @@ loss.backward()
 | Fused Linear DPO Loss           | `liger_kernel.chunked_loss.LigerFusedLinearDPOLoss`       |
 | Fused Linear ORPO Loss          | `liger_kernel.chunked_loss.LigerFusedLinearORPOLoss`      |
 | Fused Linear SimPO Loss         | `liger_kernel.chunked_loss.LigerFusedLinearSimPOLoss`     |
+| Fused Linear KTO Loss           | `liger_kernel.chunked_loss.LigerFusedLinearKTOLoss`     |
 
 ### Distillation Kernels
 
@@ -341,6 +354,7 @@ loss.backward()
 | KLDivergence                    | `liger_kernel.transformers.LigerKLDIVLoss`                  |
 | JSD                             | `liger_kernel.transformers.LigerJSD`                        |
 | Fused Linear JSD                  | `liger_kernel.transformers.LigerFusedLinearJSD`             |
+| TVD                             | `liger_kernel.transformers.LigerTVDLoss`                    |
 
 ### Experimental Kernels
 
@@ -372,7 +386,7 @@ loss.backward()
 
 - For issues, create a Github ticket in this repository
 - For open discussion, join [our discord channel](https://discord.gg/gpumode)
-- For formal collaboration, send an email to byhsu@linkedin.com
+- For formal collaboration, send an email to yannchen@linkedin.com
 
 ## Cite this work
 

{liger_kernel-0.5.3.dist-info → liger_kernel-0.5.4.dist-info}/RECORD

@@ -1,51 +1,55 @@
 liger_kernel/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 liger_kernel/env_report.py,sha256=uhdEC8OydxoZlb7B6YYcAaBF3crGFdIck-4cxaW4NJY,1728
-liger_kernel/utils.py,sha256=HJa-xVKOohDn6pLVIx-Fv0V9h0QAL3qZGQNRICI-OpI,249
+liger_kernel/utils.py,sha256=FtVUkCGBT1UNasTl6HMNycWwiwHayK6tx-ZDdA-sNX4,1884
 liger_kernel/chunked_loss/README.md,sha256=0FmkFC3hKBqyoDT5uTlIYmrvRkF-EOCR1y-EBU1LpWU,2248
-liger_kernel/chunked_loss/__init__.py,sha256=48m-8IMOAReZbi0HK5aV-KGBp2IsZSwFvdnzTNrS4bk,516
+liger_kernel/chunked_loss/__init__.py,sha256=ATu-xX5Fc49Cr6yBOGBRNTo593ZrU5ZCsIuvoIbJWw4,603
 liger_kernel/chunked_loss/cpo_loss.py,sha256=OdBR8WYdHTKpLI_c9DcuwqKSWPeAAeTyREz46Vu_cAY,3682
 liger_kernel/chunked_loss/dpo_loss.py,sha256=wgjnwzLfrMUwV5mXgrq6G1YfQKWnbiFJegaP48BGJHY,4509
 liger_kernel/chunked_loss/functional.py,sha256=THWWpCnRVhTVfnPnyvQjdBvo1JDtxhwLmtZE_yiBBqM,817
 liger_kernel/chunked_loss/fused_linear_distillation.py,sha256=5V8rdva89WyHVbmJ8JOmC4DYNOR6ByXfx3qlUieOZkI,11002
 liger_kernel/chunked_loss/fused_linear_preference.py,sha256=idK9V9NivoVITqVpiG0fEGUHSvinYWkn9-EYXZjR-KQ,18356
+liger_kernel/chunked_loss/fused_linear_rlhf.py,sha256=sAApL4GQ3YL2F-ymIAF61GCpFfBgFcWF5LB4Gzd7LgY,8044
 liger_kernel/chunked_loss/fused_linear_unpaired_preference.py,sha256=ZqYlXXhIphkJPxOS7iI70avgrr6x0skEtgpckZTYau0,9819
+liger_kernel/chunked_loss/grpo_loss.py,sha256=M5qlQR-v5Rh8N3P3dPGNhOKygDFJ4516_rJaVPzU_-c,4980
 liger_kernel/chunked_loss/jsd_loss.py,sha256=yRCQdvd3ruTWP4A_BfU8VcZ6LepSUfO0Ob7stGnueQY,6052
-liger_kernel/chunked_loss/kto_loss.py,sha256=eVNW6HVCAm32shpfhbRlk92Flnjd7G32v0gK9DUUSOQ,5655
+liger_kernel/chunked_loss/kto_loss.py,sha256=b3ffJyk97e-6XdXd4HFrYyx8wW4A-CU4gOaJSimKLtA,5476
 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=6roQjioyg-9O2qLPV8nL4U0-5UH80tdzOMTWwjvDnn8,7961
 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/tvd.py,sha256=FHJtLQI95ijqgg9UtaHpMAjSCiPxB6CduPwPMcGxelc,6405
+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/__init__.py,sha256=6v_VcV1GQ9ISgNCd-ZxtmEg_s5GTBQ9F-s1KrFkYzPQ,2265
 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/functional.py,sha256=lDOjch622dJIc78K3ePFK_H1DX00GC5kKjodjcbEgbM,4624
-liger_kernel/transformers/fused_linear_cross_entropy.py,sha256=ygU7cycCHWvSrrTgn2TseN8t-Qwfer4V7ldwhZ1E6_g,1776
+liger_kernel/transformers/cross_entropy.py,sha256=z3KTWQnFxr_IZaVjtYt0ZNEWQdDdYThN35xWkHlDGH0,1683
+liger_kernel/transformers/functional.py,sha256=ShLD3eb--XKNtllznCrOYTbo4f-1KVwzi0KLMICdrn4,4942
+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=g3i3q5McBg23A3Mnviw-Eb32le1hvN7jByzONa9ngcs,44000
 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
 liger_kernel/transformers/swiglu.py,sha256=i9WTqcNRqReU4XJs391IPbl-I5X0wG4T72D4pqGFfJg,2422
 liger_kernel/transformers/trainer_integration.py,sha256=W3ON51O5GkyzNJsItz0y5rKx-uy2f2cFfveZpqbUdhw,123
+liger_kernel/transformers/tvd.py,sha256=XrRfyJIqN6HFxXk8MYyFVZM1OLz3mtSbRZvWfZ_JerQ,450
 liger_kernel/transformers/experimental/embedding.py,sha256=2P0QYdlFyFrG5OqTzTa1wcRgDSyjBMv5i1a7BrDPDQw,881
 liger_kernel/transformers/model/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 liger_kernel/transformers/model/gemma.py,sha256=ky89b3aWPaeTGRMC-745KgixtQIRXzNAiCORAMLn9yo,9654
@@ -54,6 +58,7 @@ liger_kernel/transformers/model/llama.py,sha256=3LJFXKFDKvEakaWPc_NicSFst4Y_hdSM
 liger_kernel/transformers/model/mistral.py,sha256=MVRksI5_j_8WJu8znOHKCdSI5jSu-S7cdFYzt9m_vIQ,5180
 liger_kernel/transformers/model/mixtral.py,sha256=jpZJkpl625Q-JHWarj2MqT5mRaSsiCtg0c9vVyvOdCY,11430
 liger_kernel/transformers/model/mllama.py,sha256=qWexBdskuN3gPJvPUwt4J0nU675tGD6W7wxgRZ9Bifg,11145
+liger_kernel/transformers/model/olmo2.py,sha256=yyksS6E4fuWd8asEW8rEDBKqZpFmP4ITCM_bjIDZaoY,5124
 liger_kernel/transformers/model/phi3.py,sha256=biRa8fph9qdnQmkD9I21t5XIjpIt1i6UKU4uk8Up8pU,10292
 liger_kernel/transformers/model/qwen2.py,sha256=14UuPjxB-tjqWn85Tn4fqBFvVhVsth5iPEt8kJSMiew,9581
 liger_kernel/transformers/model/qwen2_vl.py,sha256=yMLqsfSYcvhClUpTUjGoADiOxfLB2B8240VdrPP0c8s,9851
@@ -61,9 +66,9 @@ 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-0.5.3.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
-liger_kernel-0.5.3.dist-info/METADATA,sha256=R1EvFj3Whda7t4SB6FCZn0QOVuqsbNOELuMV7j3rG7k,21323
-liger_kernel-0.5.3.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
-liger_kernel-0.5.3.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-liger_kernel-0.5.3.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
-liger_kernel-0.5.3.dist-info/RECORD,,
+liger_kernel-0.5.4.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
+liger_kernel-0.5.4.dist-info/METADATA,sha256=Zw7n3Ey6vUed4E54H9-TzKmhuOpd9P2ZFMVL-zYUnew,22255
+liger_kernel-0.5.4.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
+liger_kernel-0.5.4.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+liger_kernel-0.5.4.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
+liger_kernel-0.5.4.dist-info/RECORD,,