cumulative offsets

src/axolotl/kernels/moe/torch_grouped.py

@@ -64,15 +64,18 @@ def _call_grouped_mm(
         As2 = [a.to(dtype).contiguous().view(a.shape[0], a.shape[1]) for a in As]
         Bs2 = [b.to(dtype).contiguous().view(b.shape[0], b.shape[1]) for b in Bs]
         device = As2[0].device
-        offs = torch.tensor([a.shape[0] for a in As2], device=device, dtype=torch.int32)
+        lengths = torch.tensor(
+            [a.shape[0] for a in As2], device=device, dtype=torch.int32
+        )
+        offsets = torch.cumsum(lengths, dim=0)
         Y_cat = torch.ops.aten._grouped_mm(
-            torch.cat(As2, dim=0), torch.stack(Bs2, dim=0), offs
+            torch.cat(As2, dim=0), torch.stack(Bs2, dim=0), offsets
         )
         outs: List[torch.Tensor] = []
         start = 0
-        for m in offs.tolist():
+        for size in lengths.tolist():
-            outs.append(Y_cat[start : start + m])
+            outs.append(Y_cat[start : start + size])
-            start += m
+            start += size
         return outs
     except RuntimeError as err:
         _LOGGER.warning("torch_grouped: _grouped_mm failed (%s)", err)