Fix trl version to 0.2.12 and add Python venv instructions #143
Conversation
…ions for setting up Python virtual environment.
| "id": "ba1788d9", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "Once we have access to the TPU VM, we can setup a virtual Python environment to install our packages in:" |
There was a problem hiding this comment.
this could be an optional step. Could you rename it into something like: "While not strictly necessary, you can do an extra step to isolate your python environment by installing python-venv:" (or something like this).
| "pip install -e . -f https://storage.googleapis.com/libtpu-releases/index.html\n", | ||
| "# Install TRL and PEFT for training (see later how they are used)\n", | ||
| "pip install trl peft\n", | ||
| "pip install trl==0.2.12 peft\n", |
There was a problem hiding this comment.
What is the problem with trl 0.2.13? Can you open an issue describing the issue you observed so that we can fix it in the future?
There was a problem hiding this comment.
I noticed the same issue as @duncantech. Trl 0.2.13 as some breaking changes and so the interface is different. We need to adapt the examples to make them work with the latest version. I think this could be done in another PR tho
There was a problem hiding this comment.
@baptistecolle can you open an issue to do that later?
There was a problem hiding this comment.
LGTM thanks for pinning the trl dependency. Btw the llama fine-tuning example as the same problem and pinning dependency would be best there too https://github.com/huggingface/optimum-tpu/blob/main/examples/language-modeling/llama_tuning.ipynb
if this is not done in this PR i can open a small one to fix this
3 participants
What does this PR do?
Fixes # (issue)
Before submitting