Update base_blockwise_quantization.py (#296)

ModelTC · Jan 14, 2025 · 9943e79 · 9943e79
1 parent 74387fe
commit 9943e79
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/llmc/compression/quantization/base_blockwise_quantization.py b/llmc/compression/quantization/base_blockwise_quantization.py
@@ -840,7 +840,7 @@ def scaling_input(self, x, scales, is_gqa):
                 batch_scale = scales_tmp.view(1, -1)
                 x_tmp[i] = batch / batch_scale
         else:
-            x_tmp = x / scales.view(1, -1)
+            x_tmp = x / scales_tmp.view(1, -1)
         return x_tmp
 
     @torch.no_grad()