[Kernel] DeepEP dispatch-combine kernel integration (#18434)

Signed-off-by: Varun <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
This commit is contained in:
Varun Sundar Rabindranath
2025-06-03 12:30:02 -07:00
committed by GitHub
co-authored by Varun Sundar Rabindranath
parent 01eee40536
commit fa98d77773
23 changed files with 1950 additions and 122 deletions
@@ -18,8 +18,8 @@ def _resize_cache(x: torch.Tensor, v: tuple[int, ...]) -> torch.Tensor:
Shrink the given tensor and apply the given view to it. This is
used to resize the intermediate fused_moe caches.
"""
assert prod(
v) <= x.numel(), f"{prod(v)} <= {x.numel()}" # CUDAGRAPH unfriendly?
assert prod(v) <= x.numel(
), f"{v} ({prod(v)}) <= {x.shape} ({x.numel()})" # CUDAGRAPH unfriendly?
return x.flatten()[:prod(v)].view(*v)