I wrote a code to train a NER model before few months and it was working well. However, the same code now gives me this error:
ImportError: Using the `Trainer` with `PyTorch` requires `accelerate`: Run `pip install --upgrade accelerate`
I tried to install accelerate by following the instruction here however it is not running, this is the output when I tried to run accelerate:
2023-05-15 07:28:08.310904: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
Copy-and-paste the text below in your GitHub issue
- `Accelerate` version: 0.19.0
- Platform: Linux-5.15.107+-x86_64-with-glibc2.31
- Python version: 3.10.11
- Numpy version: 1.22.4
- PyTorch version (GPU?): 2.0.0+cu118 (True)
- System RAM: 12.68 GB
- GPU type: Tesla T4
- `Accelerate` default config:
- compute_environment: LOCAL_MACHINE
- distributed_type: NO
- mixed_precision: fp16
- use_cpu: False
- num_processes: 1
- machine_rank: 0
- num_machines: 1
- rdzv_backend: static
- same_network: False
- main_training_function: main
- downcast_bf16: False
- tpu_use_cluster: False
- tpu_use_sudo: False
I also installed TensorRT but this didn't solve the issue.
Here is my complete code for training.
Either installing !pip install accelerate -U
and then restarting the runtime, or downgrading the PyTorch to 1.3.1 version should work.
! pip install torch==1.13.1