addmm_

Author: justheuristic

bitsandbytes/autograd/_functions.py

@@ -318,7 +318,7 @@ def forward(ctx, A, B, out=None, bias=None, state=MatmulLtState()):
 
         # 4. Mixed-precision decomposition matmul
         if coo_tensorA is not None and subA is not None:
-            output += torch.matmul(subA, state.subB)
+            output.addmm_(output, subA, state.subB)
 
         # 5. Save state
         ctx.state = state