-
Notifications
You must be signed in to change notification settings - Fork 161
Open
Description
Problem Description
Hi,
The following line fails
aiter/aiter/jit/utils/cpp_extension.py
Line 467 in 6d0beab
| val = getattr(torch._C, f"_PYBIND11_{name}") |
when using vllm + aiter + gpt-oss, see below. You can also reproduce with: getattr(torch._C, "_PYBIND11_COMPILER_TYPE" using latest pytorch+rocm release at https://pytorch.org/get-started/locally/
(base) root@felix-marty-job-torch-vllm-1-rmfqs:/shared_volume/repos# export VLLM_ROCM_USE_AITER=1
export VLLM_USE_AITER_UNIFIED_ATTENTION=1
export VLLM_ROCM_USE_AITER_MHA=0
(base) root@felix-marty-job-torch-vllm-1-rmfqs:/shared_volume/repos# ^C
(base) root@felix-marty-job-torch-vllm-1-rmfqs:/shared_volume/repos# vllm serve /models/openai_gpt-oss-20b --tensor-parallel
-size 1 --enforce-eager
(APIServer pid=93425) INFO 11-12 15:30:30 [api_server.py:1897] vLLM API server version 0.11.1rc7.dev67+g1761dea1a
(APIServer pid=93425) INFO 11-12 15:30:30 [utils.py:253] non-default args: {'model_tag': '/models/openai_gpt-oss-20b', 'model': '/models/openai_gpt-oss-20b', 'enforce_eager': True}
(APIServer pid=93425) INFO 11-12 15:30:30 [model.py:630] Resolved architecture: GptOssForCausalLM
(APIServer pid=93425) ERROR 11-12 15:30:30 [config.py:306] Error retrieving safetensors: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/models/openai_gpt-oss-20b'. Use `repo_type` argument if needed., retrying 1 of 2
(APIServer pid=93425) ERROR 11-12 15:30:32 [config.py:304] Error retrieving safetensors: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/models/openai_gpt-oss-20b'. Use `repo_type` argument if needed.
(APIServer pid=93425) INFO 11-12 15:30:32 [model.py:1951] Downcasting torch.float32 to torch.bfloat16.
(APIServer pid=93425) INFO 11-12 15:30:32 [model.py:1728] Using max model len 131072
(APIServer pid=93425) [aiter] start build [module_aiter_enum] under /shared_volume/repos/aiter/aiter/jit/build/module_aiter_enum
(APIServer pid=93425) [2025-11-12 15:31:03] INFO core.py:270: start build [module_aiter_enum] under /shared_volume/repos/aiter/aiter/jit/build/module_aiter_enum
(APIServer pid=93425) Successfully preprocessed all matching files.
(APIServer pid=93425) [aiter] failed build jit [module_aiter_enum]↓↓↓↓↓↓↓↓↓↓
(APIServer pid=93425) -->[History]: Traceback (most recent call last):
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/core.py", line 513, in wrapper
(APIServer pid=93425) module = get_module(md)
(APIServer pid=93425) ^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/core.py", line 221, in get_module
(APIServer pid=93425) __mds[md_name] = importlib.import_module(f"{__package__}.{md_name}")
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "/root/miniforge3/lib/python3.12/importlib/__init__.py", line 90, in import_module
(APIServer pid=93425) return _bootstrap._gcd_import(name[level:], package, level)
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "<frozen importlib._bootstrap>", line 1387, in _gcd_import
(APIServer pid=93425) --> File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
(APIServer pid=93425) --> File "<frozen importlib._bootstrap>", line 1324, in _find_and_load_unlocked
(APIServer pid=93425) -->ModuleNotFounderror: No module named 'aiter.jit.module_aiter_enum'
(APIServer pid=93425) -->
(APIServer pid=93425) During handling of the above exception, another exception occurred:
(APIServer pid=93425)
(APIServer pid=93425) -->Traceback (most recent call last):
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/core.py", line 363, in MainFunc
(APIServer pid=93425) _jit_compile(
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1223, in _jit_compile
(APIServer pid=93425) _write_ninja_file_and_build_library(
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1333, in _write_ninja_file_and_build_library
(APIServer pid=93425) _write_ninja_file_to_build_library(
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1547, in _write_ninja_file_to_build_library
(APIServer pid=93425) common_cflags += [f"{x}" for x in _get_pybind11_abi_build_flags()]
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1086, in _get_pybind11_abi_build_flags
(APIServer pid=93425) pval = getattr(torch._C, f"_PYBIND11_{pname}")
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) -->Attributeerror: module 'torch._C' has no attribute '_PYBIND11_COMPILER_TYPE'
(APIServer pid=93425) failed build jit [module_aiter_enum]↑↑↑↑↑↑↑↑↑↑
(APIServer pid=93425) [2025-11-12 15:31:14] ERROR core.py:386: failed build jit [module_aiter_enum]↓↓↓↓↓↓↓↓↓↓
(APIServer pid=93425) -->[History]: Traceback (most recent call last):
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/core.py", line 513, in wrapper
(APIServer pid=93425) module = get_module(md)
(APIServer pid=93425) ^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/core.py", line 221, in get_module
(APIServer pid=93425) __mds[md_name] = importlib.import_module(f"{__package__}.{md_name}")
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "/root/miniforge3/lib/python3.12/importlib/__init__.py", line 90, in import_module
(APIServer pid=93425) return _bootstrap._gcd_import(name[level:], package, level)
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "<frozen importlib._bootstrap>", line 1387, in _gcd_import
(APIServer pid=93425) --> File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
(APIServer pid=93425) --> File "<frozen importlib._bootstrap>", line 1324, in _find_and_load_unlocked
(APIServer pid=93425) -->ModuleNotFounderror: No module named 'aiter.jit.module_aiter_enum'
(APIServer pid=93425) -->
(APIServer pid=93425) During handling of the above exception, another exception occurred:
(APIServer pid=93425)
(APIServer pid=93425) -->Traceback (most recent call last):
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/core.py", line 363, in MainFunc
(APIServer pid=93425) _jit_compile(
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1223, in _jit_compile
(APIServer pid=93425) _write_ninja_file_and_build_library(
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1333, in _write_ninja_file_and_build_library
(APIServer pid=93425) _write_ninja_file_to_build_library(
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1547, in _write_ninja_file_to_build_library
(APIServer pid=93425) common_cflags += [f"{x}" for x in _get_pybind11_abi_build_flags()]
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) --> File "/shared_volume/repos/aiter/aiter/jit/utils/cpp_extension.py", line 1086, in _get_pybind11_abi_build_flags
(APIServer pid=93425) pval = getattr(torch._C, f"_PYBIND11_{pname}")
(APIServer pid=93425) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=93425) -->Attributeerror: module 'torch._C' has no attribute '_PYBIND11_COMPILER_TYPE'
(APIServer pid=93425) failed build jit [module_aiter_enum]↑↑↑↑↑↑↑↑↑↑
(APIServer pid=93425) [aiter] finish build [module_aiter_enum], cost 41.49313282s
(APIServer pid=93425) [2025-11-12 15:31:14] INFO core.py:399: finish build [module_aiter_enum], cost 41.49313282s
using the recommended AITER_BRANCH: https://github.com/vllm-project/vllm/blob/54aecd9ed55d65d1f7be9d14e4e82751663438c8/docker/Dockerfile.rocm_base#L10-L11
Operating System
Ubuntu 24.04
CPU
AMD EPYC 9575F 64-Core Processor
GPU
AMD Instinct MI325X
ROCm Version
6.4.4
ROCm Component
No response
Steps to Reproduce
No response
(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support
No response
Additional Information
No response
Metadata
Metadata
Assignees
Labels
No labels