Add a test for qmatmul. (#459)

candle-core/tests/quantized_tests.rs

@@ -29,5 +29,18 @@ fn quantized_matmul() -> Result<()> {
             [343392.0, 996704.0, 1650016.0, 2303328.0]
         ]
     );
+
+    let qtensor = quantized::QTensor::new(rhs_t, (64, 4));
+    let op = quantized::QMatMul::new(std::sync::Arc::new(qtensor));
+    let res = tensor_lhs.custom_op1(op)?;
+    assert_eq!(
+        res.to_vec2::<f32>()?,
+        &[
+            [85120.43, 214561.61, 345454.9, 474748.1],
+            [213474.94, 604465.25, 1000686.4, 1388317.3],
+            [341875.88, 994283.0, 1655708.8, 2301518.3]
+        ]
+    );
+
     Ok(())
 }