From a27ac6175eaaf19e1a535e2dc8452cf62b3c8d90 Mon Sep 17 00:00:00 2001
From: Daniel Han-Chen <danielhanchen@gmail.com>
Date: Sat, 27 Jan 2024 04:03:47 +1100
Subject: [PATCH] Update utils.py

---
 unsloth/kernels/utils.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/unsloth/kernels/utils.py b/unsloth/kernels/utils.py
index 7c693a6c4e..e22e3a11b9 100644
--- a/unsloth/kernels/utils.py
+++ b/unsloth/kernels/utils.py
@@ -179,7 +179,10 @@ pass
 
 def fast_linear_forward(proj, X, temp_lora = None, out = None):
     W, W_quant, lora_A, lora_B, lora_S = get_lora_parameters(proj)
-    out = fast_gemv(X, W, W_quant, out = out)
+    if W_quant is None:
+        out = torch.matmul(X, W.t())
+    else:
+        out = fast_gemv(X, W, W_quant, out = out)
     if lora_A is not None:
 
         # Save LoRAs for inference to stop data movement costs