[Feature][MoRIIO] Support hybrid WRITE transfers with DSpark (#60544)

Signed-off-by: Yichao Zhu <Yichao.Zhu@amd.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
YukioZzz
2026-10-17 04:57:58 +00:00
committed by GitHub
co-authored by OpenAI Codex tjtanaa Claude Opus 5.5
parent a0e7137c4b
commit 41b8c15809
14 changed files with 1589 additions and 88 deletions
@@ -473,9 +473,11 @@ async def stop_profile():
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--port", type=int, default=10001)
# Port the prefill/decode servers register on (their proxy_ping_port).
parser.add_argument("--service-discovery-port", type=int, default=36367)
args = parser.parse_args()
t = start_service_discovery("0.0.0.0", 36367)
t = start_service_discovery("0.0.0.0", args.service_discovery_port)
# High-concurrency hardening. Quart's app.run() uses a shallow listen
# backlog (100) and, with app.debug=True, adds per-request overhead that
# starves the single accept loop. Under a burst of ~512 simultaneous client