RuntimeError: Could not load libtorchcodec when torchcodec being installed along with torch 2.9 RC #912
Description
🐛 Describe the bug
Install transformers
git clone https://github.com/huggingface/transformers.git && cd transformers && git fetch origin && git checkout bb45d3631ec7026db04a77d33a52b31766372160 && pip install -e .[torch,testing]Install
python3 -m pip uninstall -y torch torchvision torchaudio codecarbon torchcodec
python3 -m pip install --no-cache-dir torch==2.9.0 torchvision --index-url https://download.pytorch.org/whl/test/cu126
pip install --progress-bar=off --pre torchcodec --index-url="https://download.pytorch.org/whl/nightly/cu126"What we get
torch==2.9.0+cu126
torchcodec==0.7.0.dev20250929+cu126
torchvision==0.24.0+cu126Same error if
python3 -m pip install --no-cache-dir torch==2.9.0 torchvision torchcodec --index-url https://download.pytorch.org/whl/test/cu126
torch==2.9.0+cu126
torchcodec==0.4.0+cu126
torchvision==0.24.0+cu126
No error if
python3 -m pip install torch torchvision torchcodec --index-url https://download.pytorch.org/whl/cu126
torch==2.8.0+cu126
torchcodec==0.7.0+cu126
torchvision==0.23.0+cu126
(so it should not be the issue of FFmpeg in the CI ruuner)
Run test
RUN_SLOW=1 python3 -m pytest -v tests/models/clvp/test_modeling_clvp.py::ClvpModelForConditionalGenerationTest::test_config_attn_implementation_setter
Error
================================================================================================ test session starts ================================================================================================
platform linux -- Python 3.10.12, pytest-8.4.2, pluggy-1.6.0 -- /usr/bin/python3
cachedir: .pytest_cache
hypothesis profile 'default'
rootdir: /transformers
configfile: pyproject.toml
plugins: typeguard-4.4.4, hydra-core-1.3.2, anyio-4.11.0, hypothesis-6.140.2, asyncio-1.2.0, order-1.3.0, rerunfailures-15.1, rich-0.2.0, timeout-2.4.0, xdist-3.8.0
asyncio: mode=strict, debug=False, asyncio_default_fixture_loop_scope=function, asyncio_default_test_loop_scope=function
collected 1 item
tests/models/clvp/test_modeling_clvp.py::ClvpModelForConditionalGenerationTest::test_config_attn_implementation_setter FAILED [100%]
===================================================================================================== FAILURES ======================================================================================================
___________________________________________________________________ ClvpModelForConditionalGenerationTest.test_config_attn_implementation_setter ____________________________________________________________________
self = <tests.models.clvp.test_modeling_clvp.ClvpModelForConditionalGenerationTest testMethod=test_config_attn_implementation_setter>
def test_config_attn_implementation_setter(self):
> config, _ = self.model_tester.prepare_config_and_inputs_for_common()
tests/test_modeling_common.py:4355:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
tests/models/clvp/test_modeling_clvp.py:398: in prepare_config_and_inputs_for_common
config_and_inputs = self.prepare_config_and_inputs()
tests/models/clvp/test_modeling_clvp.py:376: in prepare_config_and_inputs
audio = ds.sort("id")[0]["audio"]
/usr/local/lib/python3.10/dist-packages/datasets/arrow_dataset.py:2862: in __getitem__
return self._getitem(key)
/usr/local/lib/python3.10/dist-packages/datasets/arrow_dataset.py:2844: in _getitem
formatted_output = format_table(
/usr/local/lib/python3.10/dist-packages/datasets/formatting/formatting.py:658: in format_table
return formatter(pa_table, query_type=query_type)
/usr/local/lib/python3.10/dist-packages/datasets/formatting/formatting.py:411: in __call__
return self.format_row(pa_table)
/usr/local/lib/python3.10/dist-packages/datasets/formatting/formatting.py:460: in format_row
row = self.python_features_decoder.decode_row(row)
/usr/local/lib/python3.10/dist-packages/datasets/formatting/formatting.py:224: in decode_row
return self.features.decode_example(row, token_per_repo_id=self.token_per_repo_id) if self.features else row
/usr/local/lib/python3.10/dist-packages/datasets/features/features.py:2096: in decode_example
return {
/usr/local/lib/python3.10/dist-packages/datasets/features/features.py:2097: in <dictcomp>
column_name: decode_nested_example(feature, value, token_per_repo_id=token_per_repo_id)
/usr/local/lib/python3.10/dist-packages/datasets/features/features.py:1409: in decode_nested_example
return schema.decode_example(obj, token_per_repo_id=token_per_repo_id) if obj is not None else None
/usr/local/lib/python3.10/dist-packages/datasets/features/audio.py:179: in decode_example
from ._torchcodec import AudioDecoder
/usr/local/lib/python3.10/dist-packages/datasets/features/_torchcodec.py:2: in <module>
from torchcodec.decoders import AudioDecoder as _AudioDecoder
/usr/local/lib/python3.10/dist-packages/torchcodec/__init__.py:10: in <module>
from . import decoders, samplers # noqa
/usr/local/lib/python3.10/dist-packages/torchcodec/decoders/__init__.py:7: in <module>
from .._core import AudioStreamMetadata, VideoStreamMetadata
/usr/local/lib/python3.10/dist-packages/torchcodec/_core/__init__.py:8: in <module>
from ._metadata import (
/usr/local/lib/python3.10/dist-packages/torchcodec/_core/_metadata.py:16: in <module>
from torchcodec._core.ops import (
/usr/local/lib/python3.10/dist-packages/torchcodec/_core/ops.py:84: in <module>
load_torchcodec_shared_libraries()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
def load_torchcodec_shared_libraries():
# Successively try to load the shared libraries for each version of FFmpeg
# that we support. We always start with the highest version, working our way
# down to the lowest version. Once we can load ALL shared libraries for a
# version of FFmpeg, we have succeeded and we stop.
#
# Note that we use two different methods for loading shared libraries:
#
# 1. torch.ops.load_library(): For PyTorch custom ops and the C++ only
# libraries the custom ops depend on. Loading libraries through PyTorch
# registers the custom ops with PyTorch's runtime and the ops can be
# accessed through torch.ops after loading.
#
# 2. importlib: For pybind11 modules. We load them dynamically, rather
# than using a plain import statement. A plain import statement only
# works when the module name and file name match exactly. Our shared
# libraries do not meet those conditions.
exceptions = []
for ffmpeg_major_version in (7, 6, 5, 4):
pybind_ops_module_name = _get_pybind_ops_module_name(ffmpeg_major_version)
decoder_library_name = f"libtorchcodec_core{ffmpeg_major_version}"
custom_ops_library_name = f"libtorchcodec_custom_ops{ffmpeg_major_version}"
pybind_ops_library_name = f"libtorchcodec_pybind_ops{ffmpeg_major_version}"
try:
torch.ops.load_library(_get_extension_path(decoder_library_name))
torch.ops.load_library(_get_extension_path(custom_ops_library_name))
pybind_ops_library_path = _get_extension_path(pybind_ops_library_name)
global _pybind_ops
_pybind_ops = _load_pybind11_module(
pybind_ops_module_name, pybind_ops_library_path
)
return
except Exception as e:
# TODO: recording and reporting exceptions this way is OK for now as it's just for debugging,
# but we should probably handle that via a proper logging mechanism.
exceptions.append((ffmpeg_major_version, e))
traceback = (
"\n[start of libtorchcodec loading traceback]\n"
+ "\n".join(f"FFmpeg version {v}: {str(e)}" for v, e in exceptions)
+ "\n[end of libtorchcodec loading traceback]."
)
> raise RuntimeError(
f"""Could not load libtorchcodec. Likely causes:
1. FFmpeg is not properly installed in your environment. We support
versions 4, 5, 6 and 7.
2. The PyTorch version ({torch.__version__}) is not compatible with
this version of TorchCodec. Refer to the version compatibility
table:
https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec.
3. Another runtime dependency; see exceptions below.
The following exceptions were raised as we tried to load libtorchcodec:
"""
f"{traceback}"
)
E RuntimeError: Could not load libtorchcodec. Likely causes:
E 1. FFmpeg is not properly installed in your environment. We support
E versions 4, 5, 6 and 7.
E 2. The PyTorch version (2.9.0+cu126) is not compatible with
E this version of TorchCodec. Refer to the version compatibility
E table:
E https://github.com/pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec.
E 3. Another runtime dependency; see exceptions below.
E The following exceptions were raised as we tried to load libtorchcodec:
E
E [start of libtorchcodec loading traceback]
E FFmpeg version 7: Could not load this library: /usr/local/lib/python3.10/dist-packages/torchcodec/libtorchcodec_core7.so
E FFmpeg version 6: Could not load this library: /usr/local/lib/python3.10/dist-packages/torchcodec/libtorchcodec_core6.so
E FFmpeg version 5: Could not load this library: /usr/local/lib/python3.10/dist-packages/torchcodec/libtorchcodec_core5.so
E FFmpeg version 4: Could not load this library: /usr/local/lib/python3.10/dist-packages/torchcodec/libtorchcodec_core4.so
E [end of libtorchcodec loading traceback].
/usr/local/lib/python3.10/dist-packages/torchcodec/_core/ops.py:69: RuntimeErrorVersions
Collecting environment information...
PyTorch version: 2.9.0+cu126
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.35
Python version: 3.10.12 (main, Aug 15 2025, 14:32:43) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.10.242-239.961.amzn2.x86_64-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to:
GPU models and configuration: GPU 0: NVIDIA A10G
Nvidia driver version: 570.172.08
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.3.0
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 48 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 16
On-line CPU(s) list: 0-15
Vendor ID: AuthenticAMD
Model name: AMD EPYC 7R32
CPU family: 23
Model: 49
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 1
Stepping: 0
BogoMIPS: 5599.85
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl
nonstop_tsc cpuid extd_apicid aperfmperf tsc_known_freq pni pclmulqdq ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch topoext ssbd ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 clzero xsaveerptr rdpru wbnoinvd arat npt nrip_save rdpid
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 256 KiB (8 instances)
L1i cache: 256 KiB (8 instances)
L2 cache: 4 MiB (8 instances)
L3 cache: 32 MiB (2 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-15
Vulnerability Gather data sampling: Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection
Vulnerability Spec rstack overflow: Mitigation; safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, IBPB conditional, STIBP always-on, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsa: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] mypy_extensions==1.1.0
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] onnx==1.19.0
[pip3] onnxruntime==1.23.0
[pip3] onnxruntime-tools==1.7.0
[pip3] torch==2.9.0+cu126
[pip3] torchcodec==0.7.0.dev20250929+cu126
[pip3] torchvision==0.24.0+cu126
[pip3] triton==3.5.0
[conda] Could not collect