Skip to content

fix bug of multicard grad ncclAllReduce #30553

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

Merged
merged 1 commit into from
Jan 19, 2021
+9 −3
Merged

fix bug of multicard grad ncclAllReduce #30553

Show file tree
Hide file tree
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: 6 additions & 2 deletions paddle/fluid/imperative/basic_engine.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -334,9 +334,13 @@ void BasicEngine::Execute() {
"Cannot find gradient of variable %s", var->Name()));
}

// leaf_accumulators_ : hooks and accumulate-grad for leaf tensor
// leaf_accumulators_ : hooks and accumulate-grad for leaf tensor,
// it should be orderly and not reapeated.
if (var->IsLeafGrad()) {
leaf_accumulators_.insert(iter->second.get());
if (std::find(leaf_accumulators_.begin(), leaf_accumulators_.end(),
iter->second.get()) == leaf_accumulators_.end()) {
leaf_accumulators_.push_back(iter->second.get());
}

if (iter->second->HasInnerVar()) {
var = iter->second->InnerVar();
Expand Down
4 changes: 3 additions & 1 deletion paddle/fluid/imperative/basic_engine.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,9 @@ class BasicEngine : public Engine {
std::vector<std::pair<GradientAccumulator*, std::shared_ptr<VariableWrapper>>>
need_accu_var_list_;
// leaf_accumulators_ is only for leaf tensor(hooks/accumulate grad)
std::unordered_set<GradientAccumulator*> leaf_accumulators_;
// It should be orderly and not repeated, because multiple cards must ensure
// that the order of vars is the same.
std::vector<GradientAccumulator*> leaf_accumulators_;

bool retain_graph_;
};
Expand Down