daniel/h3-blackwell-runtime
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Use explicit C (not x.shape[-1]) in _merge_tokens

Browse source
This commit is contained in:
Daniel Maddern 2026-08-19 22:34:16 +07:00
parent 99a7550a51
commit 571d3541e8
1 changed files with 3 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
src/h3_blackwell_runtime/qwen3vl_vision.py
View file

@ -678,11 +678,10 @@ class Qwen3VL32BVision(nn.Module):
t = int(first[0].item()) t = int(first[0].item())
h = int(first[1].item()) h = int(first[1].item())
w = int(first[2].item()) w = int(first[2].item())
C = int(x.shape[-1])
merge = 2 merge = 2
print(f"[merge_tokens] x.shape={tuple(x.shape)}, t={t}, h={h}, w={w}, " target = (t, h // merge, merge, w // merge, merge, C)
f"target=({t}, {h//merge}, {merge}, {w//merge}, {merge}, {x.shape[-1]}), " return x.view(target).permute(0, 1, 3, 2, 4, 5).reshape(-1, C // (merge * merge))
f"prod={t*(h//merge)*merge*(w//merge)*merge*x.shape[-1]}", flush=True)
return x.view((t, h // merge, merge, w // merge, merge, x.shape[-1])).permute(0, 1, 3, 2, 4, 5).reshape(-1, -1)
def forward(self, flatten_patches: torch.Tensor, grid_thw: torch.Tensor) -> tuple[torch.Tensor, list[torch.Tensor]]: def forward(self, flatten_patches: torch.Tensor, grid_thw: torch.Tensor) -> tuple[torch.Tensor, list[torch.Tensor]]:
"""Run the visual tower -> (merged, deepstack).""" """Run the visual tower -> (merged, deepstack)."""