optimize glm edge again (#12539)

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

@@ -41,6 +41,7 @@
 from ipex_llm.transformers.kv import DynamicNormalCache, DynamicFp8Cache
 from ipex_llm.transformers.models.common import merge_qkv_base
 from ipex_llm.transformers.models.utils import use_sdp, use_sdp_causal
+from ipex_llm.transformers.models.utils import make_cache_contiguous_inplaced
 from ipex_llm.transformers.models.utils import use_quantize_kv_cache, restore_fp8_kv_cache
 
 
@@ -94,7 +95,12 @@ def glm_attention_forward(
                                                         self.num_key_value_heads], dim=1)
 
     cos, sin = position_embeddings
-    query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+    if query_states.device.type == "xpu":
+        import xe_addons
+        make_cache_contiguous_inplaced(cos, sin)
+        xe_addons.rotary_two_with_cache_inplaced(query_states, key_states, cos, sin, True)
+    else:
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
 
     use_quantizekv = isinstance(past_key_value, DynamicFp8Cache)
     # sin and cos are specific to RoPE models; cache_position needed for the static cache

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

@@ -493,3 +493,11 @@ def get_q_proj_or_qkv_proj(self):
     elif hasattr(self, "qkv_proj"):
         proj = self.qkv_proj
     return proj
+
+
+def make_cache_contiguous_inplaced(cos: torch.Tensor, sin: torch.Tensor):
+    if not cos.is_contiguous():
+        new_cos = cos.contiguous()
+        new_sin = sin.contiguous()
+        cos.set_(new_cos)
+        sin.set_(new_sin)