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

[bug] Fix race condition #460

Merged
merged 1 commit into from
Sep 26, 2023
Merged

[bug] Fix race condition #460

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
14 changes: 14 additions & 0 deletions src/turbomind/models/llama/LlamaBatch.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,9 @@ void LlamaBatch<T>::verifyRequests(std::vector<std::shared_ptr<Request>>& stop_r

auto invalidate = [](const char* type, std::shared_ptr<Request>& req, int ec) {
TM_LOG_WARNING("[verifyRequests] Skipping invalid %s request for id %ld, code = %d", type, (long)req->id, ec);
// We don't need a barrier there because
// this lambda is called only for new requests
// which are visible only for rank = 0 thread.
req->signal.set_value(ec);
req.reset();
};
Expand Down Expand Up @@ -139,6 +142,12 @@ void LlamaBatch<T>::handleStopRequests(const std::vector<std::shared_ptr<Request
check_cuda_error(cudaMemsetAsync(sequence_length.getPtr<int>(), 0, sizeof(int), stream_));
check_cuda_error(cudaStreamSynchronize(stream_));
}

// When the signal is set threads from LlamaV2::forward can exit
// and free inputs/outputs tensors.
// Therefore we need to make sure that no threads from LlamaV2::internalThreadEntry
// are accessing the tensors.
llama_->shared_state_->barrier->wait();
if (rank_ == 0) {
r->signal.set_value(ec);
}
Expand Down Expand Up @@ -1112,6 +1121,11 @@ void LlamaBatch<T>::finishRequest(int index, bool force_end)
llama_->kv_cache_mgr_->update(cached_seq_[index], stream_);
}

// When the signal is set threads from LlamaV2::forward can exit
// and free inputs/outputs tensors.
// Therefore we need to make sure that no threads from LlamaV2::internalThreadEntry
// are accessing the tensors.
llama_->shared_state_->barrier->wait();
if (rank_ == 0) {
requests_[index]->signal.set_value(0);
}
Expand Down
Loading