Skip to content

[Fix] suppress remote weight loading engine w/o mooncake installed#14937

Merged
Kangyan-Zhou merged 5 commits intosgl-project:mainfrom
ZailiWang:fix-mooncake-dep
Dec 12, 2025
Merged

[Fix] suppress remote weight loading engine w/o mooncake installed#14937
Kangyan-Zhou merged 5 commits intosgl-project:mainfrom
ZailiWang:fix-mooncake-dep

Conversation

@ZailiWang
Copy link
Copy Markdown
Contributor

@ZailiWang ZailiWang commented Dec 12, 2025

Motivation

Fix the ModuleNotFoundError: No module named 'mooncake' in non-cuda CI cases introduced by #13125

Modifications

Add mooncake package check with try...except syntax. Set server_args.remote_instance_weight_loader_support_transfer_engine as False if mooncake package was not installed.

Adding mooncake installation does not work on non-CUDA platforms. Such error would be raised:

  File "/opt/.venv/lib/python3.12/site-packages/sglang/srt/managers/scheduler.py", line 2683, in run_scheduler_process
    scheduler = Scheduler(
                ^^^^^^^^^^
  File "/opt/.venv/lib/python3.12/site-packages/sglang/srt/managers/scheduler.py", line 320, in __init__
    self.tp_worker = TpModelWorker(
                     ^^^^^^^^^^^^^^
  File "/opt/.venv/lib/python3.12/site-packages/sglang/srt/managers/tp_worker.py", line 248, in __init__
    self._model_runner = ModelRunner(
                         ^^^^^^^^^^^^
  File "/opt/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 366, in __init__
    self.initialize(min_per_gpu_memory)
  File "/opt/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 451, in initialize
    register_memory_region_v2(
  File "/opt/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/remote_instance_weight_loader_utils.py", line 144, in register_memory_region_v2
    memory_snapshot = torch.cuda.memory.memory_snapshot()
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/.venv/lib/python3.12/site-packages/torch/cuda/memory.py", line 627, in memory_snapshot
    return torch._C._cuda_memorySnapshot(mempool_id)["segments"]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'torch._C' has no attribute '_cuda_memorySnapshot'

@gemini-code-assist
Copy link
Copy Markdown
Contributor

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

@ZailiWang
Copy link
Copy Markdown
Contributor Author

/tag-run-ci-label

@ZailiWang
Copy link
Copy Markdown
Contributor Author

The CI failure on Xeon was observed for #13125: https://github.com/sgl-project/sglang/actions/runs/20143344123/job/57816878503?pr=13125 but somehow merged regardlessly :(

@mingfeima
Copy link
Copy Markdown
Collaborator

@MingxuZh please check CI status.

@ZailiWang
Copy link
Copy Markdown
Contributor Author

@Kangyan-Zhou Kangyan-Zhou merged commit 4dabfbc into sgl-project:main Dec 12, 2025
56 of 88 checks passed
zhyncs added a commit that referenced this pull request Dec 12, 2025
@ZailiWang ZailiWang deleted the fix-mooncake-dep branch December 12, 2025 07:27
@ZailiWang ZailiWang mentioned this pull request Dec 12, 2025
6 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants