|
|
@@ -79,7 +79,7 @@ class RemoteMixtureOfExperts(nn.Module):
|
|
|
flat_inputs_per_expert = tuple(zip(*[tensor.split(1, dim=0) for tensor in nested_flatten(expert_inputs)]))
|
|
|
|
|
|
batch_jobs_args = tuple(
|
|
|
- (expert_logits[i], chosen_experts[i], self.k_min, self.timeout_after_k_min,
|
|
|
+ (expert_logits[i], chosen_experts[i], self.k_min, self.timeout_after_k_min, self.backward_k_min,
|
|
|
self.forward_timeout, self.backward_timeout, input_schema, *flat_inputs_per_expert[i])
|
|
|
for i in range(len(input))
|
|
|
)
|
justheuristic