liger-kernel-nightly 0.6.2.dev20251008084122__py3-none-any.whl → 0.6.2.dev20251010184612__py3-none-any.whl

liger_kernel/transformers/__init__.py

@@ -30,6 +30,7 @@ if TYPE_CHECKING:
     from liger_kernel.transformers.auto_model import AutoLigerKernelForCausalLM  # noqa: F401
     from liger_kernel.transformers.monkey_patch import _apply_liger_kernel  # noqa: F401
     from liger_kernel.transformers.monkey_patch import _apply_liger_kernel_to_instance  # noqa: F401
+    from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_falcon_h1  # noqa: F401
     from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_gemma  # noqa: F401
     from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_gemma2  # noqa: F401
     from liger_kernel.transformers.monkey_patch import apply_liger_kernel_to_gemma3  # noqa: F401
@@ -91,6 +92,7 @@ def __getattr__(name: str):
     monkey_patch_symbols = {
         "_apply_liger_kernel",
         "_apply_liger_kernel_to_instance",
+        "apply_liger_kernel_to_falcon_h1",
         "apply_liger_kernel_to_gemma",
         "apply_liger_kernel_to_gemma2",
         "apply_liger_kernel_to_gemma3",
@@ -157,6 +159,7 @@ if _TRANSFORMERS_AVAILABLE:
             "AutoLigerKernelForCausalLM",
             "_apply_liger_kernel",
             "_apply_liger_kernel_to_instance",
+            "apply_liger_kernel_to_falcon_h1",
             "apply_liger_kernel_to_gemma",
             "apply_liger_kernel_to_gemma2",
             "apply_liger_kernel_to_gemma3",

liger_kernel/transformers/monkey_patch.py

@@ -2107,6 +2107,72 @@ def apply_liger_kernel_to_internvl(
             logger.warning(f"{vision_model_name} is not supported by Liger kernel.")
 
 
+def apply_liger_kernel_to_falcon_h1(
+    rope: bool = True,
+    cross_entropy: bool = True,
+    fused_linear_cross_entropy: bool = False,
+    rms_norm: bool = True,
+    swiglu: bool = False,
+    model: PreTrainedModel = None,
+) -> None:
+    """
+    Apply Liger kernels to replace original implementation in HuggingFace Falcon-H1 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 False.
+        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.
+    """
+
+    assert not (cross_entropy and fused_linear_cross_entropy), (
+        "cross_entropy and fused_linear_cross_entropy cannot both be True."
+    )
+
+    from transformers.models.falcon_h1 import modeling_falcon_h1
+    from transformers.models.falcon_h1.modeling_falcon_h1 import FalconH1Model
+
+    if rope:
+        logger.info("Apply liger rotary pos emb.")
+        modeling_falcon_h1.apply_rotary_pos_emb = liger_rotary_pos_emb
+    if rms_norm:
+        logger.info("Apply liger RMSNorm")
+        modeling_falcon_h1.FalconH1RMSNorm = LigerRMSNorm
+    if swiglu:
+        raise NotImplementedError("LigerSwiGLUMLP is not available for Falcon-H1 models.")
+
+    if cross_entropy:
+        logger.info("Apply liger cross entropy")
+        from transformers.loss.loss_utils import nn
+
+        nn.functional.cross_entropy = liger_cross_entropy
+
+    if fused_linear_cross_entropy:
+        raise NotImplementedError("LigerFusedLinearCrossEntropy is not available for Falcon-H1 models.")
+
+    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. LlamaRMSNorm or LlamaMLP)
+
+        # get the base model from the model instance
+        base_model: FalconH1Model = getattr(model, model.base_model_prefix, model)
+
+        if rms_norm:
+            _patch_rms_norm_module(base_model.final_layernorm)
+
+        for decoder_layer in base_model.layers:
+            if swiglu:
+                _patch_swiglu_module(decoder_layer.mlp, LigerSwiGLUMLP)
+            if rms_norm:
+                _patch_rms_norm_module(decoder_layer.input_layernorm)
+                _patch_rms_norm_module(decoder_layer.pre_ff_layernorm)
+
+
 # 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,
@@ -2137,6 +2203,7 @@ MODEL_TYPE_TO_APPLY_LIGER_FN = {
     "smollm3": apply_liger_kernel_to_smollm3,
     "phi3": apply_liger_kernel_to_phi3,
     "paligemma": apply_liger_kernel_to_paligemma,
+    "falcon_h1": apply_liger_kernel_to_falcon_h1,
 }
 
 

{liger_kernel_nightly-0.6.2.dev20251008084122.dist-info → liger_kernel_nightly-0.6.2.dev20251010184612.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.6.2.dev20251008084122
+Version: 0.6.2.dev20251010184612
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 

{liger_kernel_nightly-0.6.2.dev20251008084122.dist-info → liger_kernel_nightly-0.6.2.dev20251010184612.dist-info}/RECORD

@@ -41,7 +41,7 @@ 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=1HPWpJGAZl4TmPzG5kgvNMR7g9Ifvj5usTP4x-RsJ38,8819
+liger_kernel/transformers/__init__.py,sha256=h-VQCbsM-T8l8jApA6mTsJdTnd3VeL14pUAdheruaiU,9010
 liger_kernel/transformers/auto_model.py,sha256=0qCTRZt280Bj_LcFdzo9hlaR-BWNazawXOGgoCZjgEg,1545
 liger_kernel/transformers/cross_entropy.py,sha256=z3KTWQnFxr_IZaVjtYt0ZNEWQdDdYThN35xWkHlDGH0,1683
 liger_kernel/transformers/dyt.py,sha256=i-4GPaMrl-jab9TVI5qN0-H9qycn_mCbV82ozU4nbmU,723
@@ -58,7 +58,7 @@ liger_kernel/transformers/jsd.py,sha256=DGqRnxIZxsvxo0_tbbxX3b-sDbDjC_yKufyRIHCc
 liger_kernel/transformers/kl_div.py,sha256=WLffFbh1EExD2Eb1F7lN11fo9JJC-0751WJjZAF1Fj8,409
 liger_kernel/transformers/layer_norm.py,sha256=c9pk3PEasOKYR0rhe5e5nNrnYKVCEW4VC8S6LpCq9EQ,906
 liger_kernel/transformers/llama4_rope.py,sha256=kS6PSHEwf3dS7hD7C7p8S0geugx2EMCiP0h0F7LsUoY,3639
-liger_kernel/transformers/monkey_patch.py,sha256=kGytMCLZdn-hPiNB_FhwojgBye0wmEKK6ZqsyKyCVAo,102745
+liger_kernel/transformers/monkey_patch.py,sha256=pULIVoznTTWtbXI0jJ7T1bVyLE4CKWThpmF6AwRZZk4,105681
 liger_kernel/transformers/multi_token_attention.py,sha256=K3NIY9_5TPgZ4_Rahn0xnkMXxD_fmlJHK4CWGYvGQp0,1752
 liger_kernel/transformers/qwen2vl_mrope.py,sha256=5EwSqrMdsL9MYspeBMXBsNJKvH0MOmRrtJXAJlnnlOI,1047
 liger_kernel/transformers/rms_norm.py,sha256=vkekcvTeWY8vL4H6hg3t0XeY0Ew_3OFMPHuzqlxPPVw,2719
@@ -98,9 +98,9 @@ liger_kernel/transformers/trainer/__init__.py,sha256=p7yQfklV8-467qSz_ZMimkbDF7H
 liger_kernel/transformers/trainer/orpo_trainer.py,sha256=tX0h63aOFe3rNqTmk6JpMf75UPo981yzEa6TghnjS0Q,5370
 liger_kernel/triton/__init__.py,sha256=qCiCamzCRv6lpV8IqpAc9YMdNKC7GKurClWceQPnlis,92
 liger_kernel/triton/monkey_patch.py,sha256=Rd0hUHAzDkFfHvnX7-PBaNK5EKnZhtfM_h-fgQH9HPY,1568
-liger_kernel_nightly-0.6.2.dev20251008084122.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
-liger_kernel_nightly-0.6.2.dev20251008084122.dist-info/METADATA,sha256=Uk7nue5xFD5LoCoKeAD8zHpI2-c-taOdvk1KyBdBUXE,24777
-liger_kernel_nightly-0.6.2.dev20251008084122.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
-liger_kernel_nightly-0.6.2.dev20251008084122.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-liger_kernel_nightly-0.6.2.dev20251008084122.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
-liger_kernel_nightly-0.6.2.dev20251008084122.dist-info/RECORD,,
+liger_kernel_nightly-0.6.2.dev20251010184612.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
+liger_kernel_nightly-0.6.2.dev20251010184612.dist-info/METADATA,sha256=GJBTlZW4fJ9BsrPBmV-mZlHOPm3JFqVBrsZDRLp-tNU,24777
+liger_kernel_nightly-0.6.2.dev20251010184612.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
+liger_kernel_nightly-0.6.2.dev20251010184612.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+liger_kernel_nightly-0.6.2.dev20251010184612.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
+liger_kernel_nightly-0.6.2.dev20251010184612.dist-info/RECORD,,