LLM: add min new token to perf test. (#10869)
This commit is contained in:
		
							parent
							
								
									fb2a160af3
								
							
						
					
					
						commit
						eb39c61607
					
				
					 1 changed files with 1 additions and 1 deletions
				
			
		| 
						 | 
				
			
			@ -50,7 +50,7 @@ def run_model_in_thread(model, in_out, tokenizer, result, warm_up, num_beams, in
 | 
			
		|||
    for i in range(num_trials + warm_up):
 | 
			
		||||
        st = time.perf_counter()
 | 
			
		||||
        output_ids = model.generate(input_ids, do_sample=False, max_new_tokens=out_len,
 | 
			
		||||
                                    num_beams=num_beams)
 | 
			
		||||
                                    min_new_tokens=out_len, num_beams=num_beams)
 | 
			
		||||
        torch.xpu.synchronize()
 | 
			
		||||
        end = time.perf_counter()
 | 
			
		||||
        output_ids = output_ids.cpu()
 | 
			
		||||
| 
						 | 
				
			
			
 | 
			
		|||
		Loading…
	
		Reference in a new issue