From 2f1ca04e9e44aca2a1a65a16e9ffba4da3ef71a3 Mon Sep 17 00:00:00 2001
From: erfanzar <erfanzare82@yahoo.com>
Date: Thu, 24 Oct 2024 02:19:44 +0330
Subject: [PATCH] updating `README.md`

---
 README.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/README.md b/README.md
index ac6a752..f2ee489 100644
--- a/README.md
+++ b/README.md
@@ -97,6 +97,7 @@ attention = get_cached_flash_attention(
 ### Environment Variables
 
 - `FORCE_MHA`: Set to "true", "1", or "on" to force using MHA implementation even for GQA cases
+- `FLASH_ATTN_BLOCK_PTR`: set to "1" to use `tl.make_block_ptr` for accessing pointer in fwd mode (better for H100/H200 GPUs)
 
 ## Performance Tips