optimize phi3 again: use sdp if possible (#10951)
This commit is contained in:
		
							parent
							
								
									c11170b96f
								
							
						
					
					
						commit
						aa2fa9fde1
					
				
					 1 changed files with 18 additions and 14 deletions
				
			
		| 
						 | 
				
			
			@ -39,7 +39,7 @@ from ipex_llm.transformers.models.utils import (
 | 
			
		|||
    rotate_half, should_use_fuse_rope,
 | 
			
		||||
    apply_rotary_pos_emb_cache_freq_xpu
 | 
			
		||||
)
 | 
			
		||||
from ipex_llm.transformers.models.utils import mlp_fusion_check, SILU
 | 
			
		||||
from ipex_llm.transformers.models.utils import mlp_fusion_check, SILU, use_new_esimd_sdp_fp16
 | 
			
		||||
from ipex_llm.transformers.kv import DynamicNormalCache
 | 
			
		||||
 | 
			
		||||
from typing import Optional, Tuple, List
 | 
			
		||||
| 
						 | 
				
			
			@ -93,22 +93,26 @@ def attention_forward(
 | 
			
		|||
        key_states, value_states = past_key_value.update(key_states, value_states,
 | 
			
		||||
                                                         self.layer_idx, None)
 | 
			
		||||
 | 
			
		||||
    # repeat k/v heads if n_kv_heads < n_heads
 | 
			
		||||
    key_states = repeat_kv(key_states, self.num_key_value_groups)
 | 
			
		||||
    value_states = repeat_kv(value_states, self.num_key_value_groups)
 | 
			
		||||
    if use_new_esimd_sdp_fp16(q_len, kv_seq_len, self.head_dim, query_states):
 | 
			
		||||
        import linear_q4_0
 | 
			
		||||
        attn_output = linear_q4_0.sdp_fp16(query_states, key_states, value_states, attention_mask)
 | 
			
		||||
    else:
 | 
			
		||||
        # repeat k/v heads if n_kv_heads < n_heads
 | 
			
		||||
        key_states = repeat_kv(key_states, self.num_key_value_groups)
 | 
			
		||||
        value_states = repeat_kv(value_states, self.num_key_value_groups)
 | 
			
		||||
 | 
			
		||||
    attn_weights = torch.matmul(query_states,
 | 
			
		||||
                                key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
 | 
			
		||||
        attn_weights = torch.matmul(query_states,
 | 
			
		||||
                                    key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
 | 
			
		||||
 | 
			
		||||
    if attention_mask is not None:
 | 
			
		||||
        attn_weights = attn_weights + attention_mask
 | 
			
		||||
        if attention_mask is not None:
 | 
			
		||||
            attn_weights = attn_weights + attention_mask
 | 
			
		||||
 | 
			
		||||
    # upcast attention to fp32
 | 
			
		||||
    attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1,
 | 
			
		||||
                                               dtype=torch.float32).to(value_states.dtype)
 | 
			
		||||
    attn_weights = torch.nn.functional.dropout(attn_weights, p=self.attention_dropout,
 | 
			
		||||
                                               training=self.training)
 | 
			
		||||
    attn_output = torch.matmul(attn_weights, value_states)
 | 
			
		||||
        # upcast attention to fp32
 | 
			
		||||
        attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1,
 | 
			
		||||
                                                   dtype=torch.float32).to(value_states.dtype)
 | 
			
		||||
        attn_weights = torch.nn.functional.dropout(attn_weights, p=self.attention_dropout,
 | 
			
		||||
                                                   training=self.training)
 | 
			
		||||
        attn_output = torch.matmul(attn_weights, value_states)
 | 
			
		||||
 | 
			
		||||
    attn_output = attn_output.transpose(1, 2).contiguous()
 | 
			
		||||
    attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
 | 
			
		||||
| 
						 | 
				
			
			
 | 
			
		|||
		Loading…
	
		Reference in a new issue