From 1096f88e2b8d67d65ed21ee23c6aed2e3e9756d6 Mon Sep 17 00:00:00 2001
From: Yue Wu <14996988+yue-here@users.noreply.github.com>
Date: Fri, 24 May 2024 11:14:32 -0700
Subject: [PATCH] sampling bug fix in diffusers tutorial "basic_training.md"
 (#8223)

sampling bug fix in basic_training.md

In the diffusers basic training tutorial, setting the manual seed argument (generator=torch.manual_seed(config.seed)) in the pipeline call inside evaluate() function rewinds the dataloader shuffling, leading to overfitting due to the model seeing same sequence of training examples after every evaluation call. Using generator=torch.Generator(device='cpu').manual_seed(config.seed) avoids this.
---
 docs/source/en/tutorials/basic_training.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/source/en/tutorials/basic_training.md b/docs/source/en/tutorials/basic_training.md
index c97ae2d62f..4f7c29cfe5 100644
--- a/docs/source/en/tutorials/basic_training.md
+++ b/docs/source/en/tutorials/basic_training.md
@@ -260,7 +260,7 @@ Then, you'll need a way to evaluate the model. For evaluation, you can use the [
 ...     # The default pipeline output type is `List[PIL.Image]`
 ...     images = pipeline(
 ...         batch_size=config.eval_batch_size,
-...         generator=torch.manual_seed(config.seed),
+...         generator=torch.Generator(device='cpu').manual_seed(config.seed), # Use a separate torch generator to avoid rewinding the random state of the main training loop
 ...     ).images
 
 ...     # Make a grid out of the images