From 3b3208fc061a385e9675cbbb2d0309d4eda5223e Mon Sep 17 00:00:00 2001
From: Daniel Han <danielhanchen@gmail.com>
Date: Thu, 13 Feb 2025 01:57:45 -0800
Subject: [PATCH] Update llama.py

---
 unsloth/models/llama.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 0b567b023a..d6916814ae 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -707,7 +707,7 @@ def LlamaModel_fast_forward(
     # Ignore attention_mask
     if attention_mask is None:
         padding_mask = None
-    elif self.training:
+    elif attention_mask is not None and self.training:
         attention_mask = None
         padding_mask = None
     else:
@@ -723,6 +723,7 @@ def LlamaModel_fast_forward(
             past_key_values_length,
             sliding_window = getattr(self.config, "sliding_window", None),
         )
+        attention_mask = attention_mask.to(torch.bool)
     pass
 
     hidden_states = inputs_embeds