optimize minicpm3 again (#12047)

python/llm/src/ipex_llm/transformers/convert.py

@@ -1775,16 +1775,16 @@ def _optimize_post(model, lightweight_bmm=False):
     elif model.config.model_type == "gemma2":
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)
+        from ipex_llm.transformers.models.common import mlp_silu_forward
         from ipex_llm.transformers.models.gemma import gemma_rms_norm_forward
         from ipex_llm.transformers.models.gemma2 import gemma2_attention_forward
         from ipex_llm.transformers.models.gemma2 import gemma2_model_forward
-        from ipex_llm.transformers.models.gemma2 import gemma2_mlp_forward
         from transformers.models.gemma2.modeling_gemma2 import Gemma2RMSNorm, Gemma2Attention
         from transformers.models.gemma2.modeling_gemma2 import Gemma2Model, Gemma2MLP
         convert_forward(model, Gemma2RMSNorm, gemma_rms_norm_forward)
         convert_forward(model, Gemma2Attention, gemma2_attention_forward)
         convert_forward(model, Gemma2Model, gemma2_model_forward)
-        convert_forward(model, Gemma2MLP, gemma2_mlp_forward)
+        convert_forward(model, Gemma2MLP, mlp_silu_forward)
     elif model.config.model_type == "Yi":
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)
@@ -1968,7 +1968,9 @@ def _optimize_post(model, lightweight_bmm=False):
         modeling_module_name = model.__class__.__module__
         module = importlib.import_module(modeling_module_name)
         from ipex_llm.transformers.models.common import rms_norm_forward
+        from ipex_llm.transformers.models.common import mlp_silu_forward
         convert_forward(model, module.MiniCPMRMSNorm, rms_norm_forward)
+        convert_forward(model, module.MiniCPMMLP, mlp_silu_forward)
         from ipex_llm.transformers.models.minicpm3 import minicpm3_attention_forward
         convert_forward(model, module.MiniCPMAttention, minicpm3_attention_forward)
     elif model.config.model_type == "minicpmv":

python/llm/src/ipex_llm/transformers/models/common.py

@@ -69,6 +69,16 @@ def fuse_mlp_base(module: torch.nn.Module, act: int, x: torch.Tensor):
         return module.down_proj(module.act_fn(module.gate_proj(x)) * module.up_proj(x))
 
 
+def mlp_silu_forward(self, x: torch.Tensor):
+    from ipex_llm.transformers.models.utils import SILU
+    return fuse_mlp_base(self, SILU, x)
+
+
+def mlp_gelu_forward(self, x: torch.Tensor):
+    from ipex_llm.transformers.models.utils import GELU
+    return fuse_mlp_base(self, GELU, x)
+
+
 def attention_softmax(attn_weights: torch.Tensor, training: bool):
     if attn_weights.is_contiguous() and attn_weights.device.type == "xpu" and not training:
         import xe_addons

python/llm/src/ipex_llm/transformers/models/gemma2.py

@@ -34,7 +34,7 @@
 import torch
 
 from typing import Optional, Tuple
-from ipex_llm.transformers.models.common import merge_qkv_base, fuse_mlp_base
+from ipex_llm.transformers.models.common import merge_qkv_base
 from ipex_llm.transformers.models.utils import GELU
 from ipex_llm.transformers.models.utils import should_use_fuse_rope, use_sdp, use_sdp_causal
 from transformers.cache_utils import Cache
@@ -184,7 +184,3 @@ def gemma2_attention_forward(
         attn_weights = None
 
     return attn_output, attn_weights, past_key_value
-
-
-def gemma2_mlp_forward(self, x: torch.Tensor):
-    return fuse_mlp_base(self, GELU, x)

python/llm/src/ipex_llm/transformers/models/minicpm3.py

@@ -92,8 +92,9 @@ def minicpm3_attention_forward(
             xe_addons.rotary_half_inplaced(inv_freq, position_ids,
                                            query_states[:, :, :, self.qk_nope_head_dim:],
                                            key_states[:, :, :, self.qk_nope_head_dim:])
-            query_states[:, :, :, self.qk_nope_head_dim:] *= self.rotary_emb.scaling_factor
+            if self.rotary_emb.scaling_factor != 1.0:
-            key_states[:, :, :, self.qk_nope_head_dim:] *= self.rotary_emb.scaling_factor
+                query_states[:, :, :, self.qk_nope_head_dim:] *= self.rotary_emb.scaling_factor
+                key_states[:, :, :, self.qk_nope_head_dim:] *= self.rotary_emb.scaling_factor
         else:
             invalidInputError(f"unknown rope method: {self.rotary_emb.__class__.__name__}")
     else: