From b7ad3d9d93756770046a9eec2f57502a0226c179 Mon Sep 17 00:00:00 2001 From: sgwhat Date: Fri, 22 Mar 2024 17:44:20 +0800 Subject: [PATCH] remove redundant warm-up for to optimize peformance --- modules/text_generation.py | 5 ----- 1 file changed, 5 deletions(-) diff --git a/modules/text_generation.py b/modules/text_generation.py index 289e6d533d..49dbe8027b 100644 --- a/modules/text_generation.py +++ b/modules/text_generation.py @@ -403,11 +403,6 @@ def generate_with_callback(callback=None, *args, **kwargs): def generate_with_streaming(**kwargs): return Iteratorize(generate_with_callback, [], kwargs, callback=None) - # warm-up - with torch.no_grad(): - shared.model.generate(**generate_params) - torch.xpu.synchronize() - with generate_with_streaming(**generate_params) as generator: cumulative_reply = '' starting_from = 0 if shared.is_seq2seq else len(input_ids[0])