Skip to content

WA for qwen3 moe concurrency handling#4068

Open
dtrawins wants to merge 5 commits intomainfrom
wa-qwen-moe
Open

WA for qwen3 moe concurrency handling#4068
dtrawins wants to merge 5 commits intomainfrom
wa-qwen-moe

Conversation

@dtrawins
Copy link
Collaborator

🛠 Summary

JIRA/Issue if applicable.
Describe the changes.

🧪 Checklist

  • Unit tests added.
  • The documentation updated.
  • Change follows security best practices.
    ``

Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds a server-side workaround to default MOE_USE_MICRO_GEMM_PREFILL=0 when not already set, intended to mitigate a Qwen3 MoE concurrency issue (noted as iGPU-related) during OVMS startup.

Changes:

  • Set MOE_USE_MICRO_GEMM_PREFILL to "0" at process start if the variable is not present.
  • Apply the workaround in the CLI startup path (Server::start(argc, argv)).

You can also share your feedback on Copilot code review. Take the survey.

src/server.cpp Outdated
Comment on lines +533 to +536
// This is WA for concurrency handling issue in iGPU. It is expected to be fixed in 2026.2
if (getenv("MOE_USE_MICRO_GEMM_PREFILL") == nullptr) {
setenv("MOE_USE_MICRO_GEMM_PREFILL", "0", 0);
}
src/server.cpp Outdated
Comment on lines 533 to 537
// This is WA for concurrency handling issue in iGPU. It is expected to be fixed in 2026.2
if (getenv("MOE_USE_MICRO_GEMM_PREFILL") == nullptr) {
setenv("MOE_USE_MICRO_GEMM_PREFILL", "0", 0);
}
auto paramsOrExit = parseArgs(argc, argv);
src/server.cpp Outdated
int Server::start(int argc, char** argv) {
// This is WA for concurrency handling issue in iGPU. It is expected to be fixed in 2026.2
if (getenv("MOE_USE_MICRO_GEMM_PREFILL") == nullptr) {
setenv("MOE_USE_MICRO_GEMM_PREFILL", "0", 0);
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants