From 512769a995ef7e4111f467797ebac74a1c724818 Mon Sep 17 00:00:00 2001
From: harras3 <harrasmansoor3@gmail.com>
Date: Mon, 3 Nov 2025 21:04:42 +0500
Subject: [PATCH] removed outdated warning

---
 docs/source/reducing_memory_usage.md | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/docs/source/reducing_memory_usage.md b/docs/source/reducing_memory_usage.md
index 016c493124b..f258c0a20f8 100644
--- a/docs/source/reducing_memory_usage.md
+++ b/docs/source/reducing_memory_usage.md
@@ -90,9 +90,6 @@ from trl import SFTConfig
 training_args = SFTConfig(..., packing=True, max_length=512)
 ```
 
-> [!WARNING]
-> Packing may cause batch contamination, where adjacent sequences influence one another. This can be problematic for some applications. For more details, see [#1230](https://github.com/huggingface/trl/issues/1230).
-
 ## Liger for reducing peak memory usage
 
 > [Liger Kernel](https://github.com/linkedin/Liger-Kernel) is a collection of Triton kernels designed specifically for LLM training. It can effectively increase multi-GPU training throughput by 20% and reduce memory usage by 60%.