support batch forward for q4_k, q6_k (#11325)
This commit is contained in:
		
							parent
							
								
									e8dd8e97ef
								
							
						
					
					
						commit
						8a3247ac71
					
				
					 1 changed files with 1 additions and 1 deletions
				
			
		| 
						 | 
				
			
			@ -332,7 +332,7 @@ def use_batch_forward(x: torch.Tensor, qtype: int, output_len: int):
 | 
			
		|||
        and output_len % 32 == 0
 | 
			
		||||
        and device in ["arc", "flex", "pvc", "mtl"]
 | 
			
		||||
        and qtype in [SYM_INT4, ASYM_INT4, SYM_INT8, FP4,
 | 
			
		||||
                      FP8E5, FP6, FP8E4]
 | 
			
		||||
                      FP8E5, FP6, FP8E4, Q4_K, Q6_K]
 | 
			
		||||
        and batch_size <= 64
 | 
			
		||||
    )
 | 
			
		||||
    if hard_condition:
 | 
			
		||||
| 
						 | 
				
			
			
 | 
			
		|||
		Loading…
	
		Reference in a new issue