disable linear fusion when using compile

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

Author: jiqing-feng

optimum/exporters/ipex/modeling_utils.py

@@ -229,7 +229,7 @@ def _llama_model_forward(
     input_lens = attention_mask.cumsum(-1)[:, -1].to(torch.int32)
     seq_len_tensor = torch.cat((input_lens.new_tensor([0]), input_lens.cumsum(-1).int()))
     query_len_tensor = torch.arange(seq_len_tensor.shape[0], device=device).int()
-    max_input_lens = input_lens.max().item()
+    max_input_lens = input_lens.max()
 
     if past_key_values_length == 0 and past_key_values is not None:
         # first token, remove the padding from hidden_states, varlen do not accept attention mask
@@ -357,7 +357,7 @@ def _falcon_model_forward(
     input_lens = attention_mask.cumsum(-1)[:, -1].to(torch.int32)
     seq_len_tensor = torch.cat((input_lens.new_tensor([0]), input_lens.cumsum(-1).int()))
     query_len_tensor = torch.arange(seq_len_tensor.shape[0], device=device).int()
-    max_input_lens = input_lens.max().item()
+    max_input_lens = input_lens.max()
 
     if past_key_values_length == 0 and past_key_values is not None:
         # first token, remove the padding from hidden_states, varlen do not accept attention mask
@@ -499,7 +499,7 @@ def _gpt2_model_forward(
     input_lens = attention_mask.cumsum(-1)[:, -1].to(torch.int32)
     seq_len_tensor = torch.cat((input_lens.new_tensor([0]), input_lens.cumsum(-1).int()))
     query_len_tensor = torch.arange(seq_len_tensor.shape[0], device=device).int()
-    max_input_lens = input_lens.max().item()
+    max_input_lens = input_lens.max()
 
     if past_length == 0 and past_key_values is not None:
         # first token, remove the padding from hidden_states, varlen do not accept attention mask
@@ -635,7 +635,7 @@ def _qwen2_model_forward(
     input_lens = attention_mask.cumsum(-1)[:, -1].to(torch.int32)
     seq_len_tensor = torch.cat((input_lens.new_tensor([0]), input_lens.cumsum(-1).int()))
     query_len_tensor = torch.arange(seq_len_tensor.shape[0], device=device).int()
-    max_input_lens = input_lens.max().item()
+    max_input_lens = input_lens.max()
 
     if past_key_values_length == 0 and past_key_values is not None:
         # first token, remove the padding from hidden_states, varlen do not accept attention mask
@@ -752,11 +752,11 @@ def attention_interface(
         if past_key_value is None:
             n_rep = query.shape[1] // key.shape[1]
             attn_output = torch.nn.functional.scaled_dot_product_attention(
-                query.reshape(input_lens.shape[0], input_lens.max().item(), -1, query.shape[-1]).transpose(1, 2),
-                key.reshape(input_lens.shape[0], input_lens.max().item(), -1, key.shape[-1])
+                query.reshape(input_lens.shape[0], input_lens.max(), -1, query.shape[-1]).transpose(1, 2),
+                key.reshape(input_lens.shape[0], input_lens.max(), -1, key.shape[-1])
                 .transpose(1, 2)
                 .repeat_interleave(n_rep, 1),
-                value.reshape(input_lens.shape[0], input_lens.max().item(), -1, value.shape[-1])
+                value.reshape(input_lens.shape[0], input_lens.max(), -1, value.shape[-1])
                 .transpose(1, 2)
                 .repeat_interleave(n_rep, 1),
                 attn_mask=attention_mask,
@@ -883,13 +883,11 @@ def __init__(self, module, device, config) -> None:
             self.q_slice = self.q_proj.weight.shape[0]
             self.k_slice = self.q_slice + self.k_proj.weight.shape[0]
             self.v_slice = self.k_slice + self.v_proj.weight.shape[0]
-            if self.module_device.type == "cpu":
-                if module.o_proj.__class__.__name__ not in ["LinearAllreduce"]:
+            if not config.compile and module.o_proj.__class__.__name__ not in ["LinearAllreduce"]:
+                if self.module_device.type == "cpu":
                     self.mha_linear_add = LinearAdd(module.o_proj)
-
                 elif self.module_device.type == "xpu":
-                    if module.o_proj.__class__.__name__ not in ["LinearAllreduce"]:
-                        self.mha_linear_add = XPULinearAdd(module.o_proj)
+                    self.mha_linear_add = XPULinearAdd(module.o_proj)
 
     def qkv_gemm(self, hidden_states):
         if hasattr(self, "concat_qkv"):
@@ -932,7 +930,7 @@ def __init__(self, module, device, config) -> None:
         self.num_key_value_heads = config.num_key_value_heads
         super().__init__(module, device, config)
         _setattr_from_module(self, module)
-        if getattr(config, "quantization_config", None) is None:
+        if not config.compile and getattr(config, "quantization_config", None) is None:
             self.c_attn_linear = nn.Linear(self.c_attn.weight.shape[0], self.c_attn.weight.shape[1])
             self.c_attn_linear.weight = nn.Parameter(self.c_attn.weight.t())
             self.c_attn_linear.bias = self.c_attn.bias
@@ -976,7 +974,7 @@ def __init__(self, module, device, config) -> None:
         _setattr_from_module(self, module)
         self.config = config
         self.module_device = device
-        if getattr(config, "quantization_config", None) is None:
+        if not config.compile and getattr(config, "quantization_config", None) is None:
             if self.module_device.type == "cpu":
                 # LinearAllreduce and LinearLayer cannot use fused op LinearAdd
                 if module.down_proj.__class__.__name__ not in ["LinearAllreduce"]:
@@ -1009,7 +1007,7 @@ def __init__(self, module, device, config) -> None:
         _setattr_from_module(self, module)
         self.config = config
         self.module_device = device
-        if getattr(config, "quantization_config", None) is None:
+        if not config.compile and getattr(config, "quantization_config", None) is None:
             # LinearAllreduce and LinearLayer cannot use fused op LinearAdd
             if self.module_device.type == "cpu":
                 self.linear_gelu = LinearGelu(module.dense_h_to_4h)
@@ -1049,7 +1047,7 @@ def __init__(self, module, device, config) -> None:
         self.config = config
         self.module_device = device
 
-        if getattr(config, "quantization_config", None) is None:
+        if not config.compile and getattr(config, "quantization_config", None) is None:
             self.c_fc_linear = nn.Linear(self.c_fc.weight.shape[0], self.c_fc.weight.shape[1])
             self.c_fc_linear.weight = nn.Parameter(self.c_fc.weight.t())
             self.c_fc_linear.bias = self.c_fc.bias
@@ -1058,11 +1056,8 @@ def __init__(self, module, device, config) -> None:
             self.c_proj_linear.bias = self.c_proj.bias
             if self.module_device.type == "cpu":
                 self.linear_new_gelu = LinearNewGelu(self.c_fc_linear)
-
-            if self.module_device.type == "cpu":
                 if self.c_proj_linear not in ["LinearAllreduce"]:
                     self.linear_add = LinearAdd(self.c_proj_linear)
-
             elif self.module_device.type == "xpu":
                 if self.c_proj_linear not in ["LinearAllreduce"]:
                     self.linear_add = XPULinearAdd(self.c_proj_linear)
@@ -1234,7 +1229,7 @@ def __init__(self, module, device, config):
         super().__init__()
         _setattr_from_module(self, module)
         self.module_device = device
-        if getattr(config, "quantization_config", None) is None:
+        if not config.compile and getattr(config, "quantization_config", None) is None:
             if self.module_device.type == "cpu":
                 self.linear_gelu = LinearGelu(module.dense)
             elif self.module_device.type == "xpu":

optimum/intel/ipex/modeling_base.py

@@ -135,7 +135,8 @@ def __init__(
         self.use_cache = kwargs.get("use_cache", False)
         self.model_save_dir = model_save_dir
         self._add_patch = _is_patched_with_ipex(model, self.export_feature, self.use_cache)
-        self.compiled = False
+        self.compile = self.can_compile()
+        model.config.compile = compile
 
         self.input_names = set(inspect.signature(model.forward).parameters)
 
@@ -147,9 +148,10 @@ def __init__(
         if hasattr(self.auto_model_class, "register"):
             self.auto_model_class.register(AutoConfig, self.__class__)
 
-        self.maybe_apply_torch_compile()
+        if self.compile:
+            self.apply_torch_compile()
 
-        if warmup and not self.compiled:
+        if warmup and not self.compile:
             self._init_warmup()
 
     @classmethod
@@ -220,24 +222,27 @@ def to(self, device: Union[torch.device, str]):
     def can_generate(self):
         return isinstance(self, GenerationMixin)
 
-    def maybe_apply_torch_compile(self):
+    def can_compile(self):
         if (
             self.model.device.type != "cpu"
             or self.config.model_type in _COMPILE_NOT_READY_MODEL_TYPES
             or is_ipex_version("<", _IPEX_MINIMUM_VERSION_FOR_COMPILE)
             or getattr(self.config, "quantization_config", None)
         ):
-            return
+            return False
         if self.use_cache and not self._supports_static_cache:
-            return
+            return False
+
+        return True
+
+    def apply_torch_compile(self):
         from torch._inductor import config as inductor_config
 
         # System level optimization
         inductor_config.cpp_wrapper = True
         os.environ["TORCHINDUCTOR_FREEZING"] = "1"
         logger.info("Enable torch.compile optimization")
         self.model.forward = torch.compile(self.model.forward)
-        self.compiled = True
 
     def _init_warmup(self):
         inputs = prepare_jit_inputs(self.model, self.export_feature, False)
@@ -317,7 +322,7 @@ def __init__(
         if hasattr(self.model_cls, "_convert_to_bloom_cache"):
             self._convert_to_bloom_cache = self.model_cls._convert_to_bloom_cache
 
-        if warmup and not self.compiled:
+        if warmup and not self.compile:
             self._init_warmup()
 
     @torch.no_grad()
@@ -337,7 +342,7 @@ def _prepare_generation_config(
         kwargs["use_cache"] = self.use_cache
         generation_config, model_kwargs = super()._prepare_generation_config(generation_config, **kwargs)
         generation_method = generation_config.get_generation_mode().value
-        if self.compiled and generation_config.cache_implementation != "ipex_paged" and self._supports_static_cache:
+        if self.compile and generation_config.cache_implementation != "ipex_paged" and self._supports_static_cache:
             # Use static cache for torch compile
             generation_config.cache_implementation = "static"
         if generation_method not in _IPEX_EXPORTED_GENERATION_METHODS:
@@ -448,7 +453,7 @@ def __init__(
         if hasattr(self.model_cls, "_convert_to_standard_cache"):
             self._convert_to_standard_cache = self.model_cls._convert_to_standard_cache
 
-        if warmup and not self.compiled:
+        if warmup and not self.compile:
             self._init_warmup()
 
     @torch.no_grad()
@@ -465,7 +470,7 @@ def _prepare_generation_config(
     ) -> Tuple[GenerationConfig, Dict]:
         generation_config, model_kwargs = super()._prepare_generation_config(generation_config, **kwargs)
         # Use static cache for torch.compile
-        if self.compiled:
+        if self.compile:
             generation_config.cache_implementation = "static"
 
         return generation_config, model_kwargs