From 936f6a48407ee111f52742cf48eccc61f6b62325 Mon Sep 17 00:00:00 2001
From: Laurent Mazare <laurent.mazare@gmail.com>
Date: Fri, 8 Mar 2024 23:12:13 +0100
Subject: [PATCH] Fix dequantization. (#1823)

---
 candle-core/src/quantized/mod.rs | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/candle-core/src/quantized/mod.rs b/candle-core/src/quantized/mod.rs
index f7abcd93..47307f2e 100644
--- a/candle-core/src/quantized/mod.rs
+++ b/candle-core/src/quantized/mod.rs
@@ -398,7 +398,7 @@ impl QMatMul {
             _ => DEQUANTIZE_ALL.with(|b| *b),
         };
         let t = if dequantize {
-            let tensor = qtensor.dequantize(&Device::Cpu)?;
+            let tensor = qtensor.dequantize(&qtensor.device())?;
             Self::Tensor(tensor)
         } else {
             Self::QTensor(qtensor)