From 665908eb70783b2cf8e558f0eb0913f66a5ca186 Mon Sep 17 00:00:00 2001
From: Daniel Han-Chen <danielhanchen@gmail.com>
Date: Sat, 3 Feb 2024 22:28:20 +1100
Subject: [PATCH] Update llama.py

---
 unsloth/models/llama.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 99ce923d36..a2ff511992 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -221,7 +221,7 @@ pass
 def fast_rms_layernorm_inference(self, X, temp1 = None, temp2 = None):
     old_dtype = X.dtype
 
-    if XX is None:
+    if temp1 is None:
         bsz, _, hd = X.shape
         temp1 = torch.empty((2, bsz, 1, hd), dtype = torch.float32, device = "cuda")
         temp2 = torch.empty((bsz, 1, hd), dtype = old_dtype, device = "cuda")
@@ -680,9 +680,9 @@ def LlamaModel_fast_forward_inference(
             decoder_layer.self_attn,
             hidden_states,
             past_key_values[idx],
-            temp_QA = temp_QA,
-            temp_KV = temp_KV,
-            RH_Q = RH_Q,
+            temp_QA,
+            temp_KV,
+            RH_Q,
         )
         hidden_states += residual