[FEATURE] add liger_kernel support

gptqmodel/models/auto.py

@@ -102,13 +102,15 @@ def from_pretrained(
         pretrained_model_name_or_path: str,
         quantize_config: QuantizeConfig,
         trust_remote_code: bool = False,
+        use_liger_kernel: bool = False,
         **model_init_kwargs,
     ) -> BaseGPTQModel:
         model_type = check_and_get_model_type(pretrained_model_name_or_path, trust_remote_code)
         return MODEL_MAP[model_type].from_pretrained(
             pretrained_model_name_or_path=pretrained_model_name_or_path,
             quantize_config=quantize_config,
             trust_remote_code=trust_remote_code,
+            use_liger_kernel=use_liger_kernel,
             **model_init_kwargs,
         )
 

gptqmodel/models/base.py

@@ -993,6 +993,7 @@ def from_pretrained(
         pretrained_model_name_or_path: str,
         quantize_config: QuantizeConfig,
         trust_remote_code: bool = False,
+        use_liger_kernel: bool = False,
         torch_dtype: [str | torch.dtype] = "auto",
         **model_init_kwargs,
     ):
@@ -1031,6 +1032,15 @@ def skip(*args, **kwargs):
 
         config = AutoConfig.from_pretrained(pretrained_model_name_or_path, **model_init_kwargs)
 
+        if use_liger_kernel:
+            from liger_kernel.transformers.monkey_patch import MODEL_TYPE_TO_APPLY_LIGER_FN
+
+            apply_fn = MODEL_TYPE_TO_APPLY_LIGER_FN.get(config.model_type, None)
+            if apply_fn is None:
+                raise ValueError(f"apply_fn is not defined for model type {config.model_type}")
+
+            apply_fn()
+
         if torch_dtype == "auto":
             torch_dtype = auto_dtype_from_config(config)
         elif not isinstance(torch_dtype, torch.dtype):

requirements.txt

@@ -17,3 +17,4 @@ intel_extension_for_transformers>=1.4.2
 auto-round==0.3
 huggingface-hub>=0.24.2
 lm_eval==0.4.3
+liger-kernel>=0.3.0