Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix RemoteMixtureOfExperts and RemoteSwitchMixtureOfExperts backward() on GPU #626

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
+5 −3
Open

Fix RemoteMixtureOfExperts and RemoteSwitchMixtureOfExperts backward() on GPU #626

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions hivemind/moe/client/moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -309,16 +309,18 @@ def backward(cls, ctx, *raw_grads: torch.Tensor) -> Tuple[Optional[torch.Tensor]

num_samples, max_experts = dummy_grad_mask.shape

inputs_per_expert = zip(*(tensor[alive_ii].split(1, dim=0) for tensor in flat_inputs_cpu))
alive_ii_cpu = alive_ii.cpu()
alive_jj_cpu = alive_jj.cpu()
inputs_per_expert = zip(*(tensor[alive_ii_cpu].split(1, dim=0) for tensor in flat_inputs_cpu))
grad_outputs_per_expert = zip(
*(tensor[alive_ii, alive_jj].split(1, dim=0) for tensor in flat_grad_outputs_cpu)
*(tensor[alive_ii_cpu, alive_jj_cpu].split(1, dim=0) for tensor in flat_grad_outputs_cpu)
)
backward_schema = tuple(nested_flatten((info["forward_schema"], info["outputs_schema"])))

# dispatch tasks to all remote experts, collect responses
pending_tasks = {}
for i, j, inputs_ij, grad_outputs_ij in zip(
alive_ii.cpu().numpy(), alive_jj.cpu().numpy(), inputs_per_expert, grad_outputs_per_expert
alive_ii_cpu.numpy(), alive_jj_cpu.numpy(), inputs_per_expert, grad_outputs_per_expert
):
expert: RemoteExpert = expert_per_sample[i.item()][j.item()]
stub = get_server_stub(expert.p2p, expert.peer_id)
Expand Down
Loading