transformers 5.3.0 requires accelerate >= 1.1.0
- Dominant language
- Python
- Stars
- 2k
- Forks
- 561
- Avg merge
- 4d 5h
- Merged PRs (30d)
- 145
Description
**Describe the bug**
NemoRL `pyproject.toml` has `transformers==5.3.0` and `accelerate>=0.26`. However this version of transformers requires `accelerate>=1.1.0`.
Caught this crash because of it:
```
[36m(XXXWorker pid=1783492)[0m File "/opt/ray_venvs/nemo_rl.models.policy.workers.XXX/lib/python3.12/site-packages/accelerate/big_modeling.py", line 135, in register_empty_parameter[32m [repeated 6x across cluster][0m
[36m(XXXWorker pid=1783492)[0m module._parameters[name] = param_cls(module._parameters[name].to(device), **kwargs)[32m [repeated 6x across cluster][0m
[36m(XXXWorker pid=1783492)[0m ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^[32m [repeated 6x across cluster][0m
[36m(XXXWorker pid=1783492)[0m TypeError: Parameter.__new__() got an unexpected keyword argument '_is_hf_initialized'[32m [repeated 6x across cluster][0m
[36m(XXXWorker pid=1783492)[0m [rank6]:[W406 20:06:50.993913153 ProcessGroupNCCL.cpp:1553] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())[32m [repeated 6x across cluster][0m
```
Contributor guide
Assessment
This issue has not been assessed yet.