make method comparison device agnostic, so it can expand to more accelerators like XPU #2610
Conversation
accelerators like XPU Signed-off-by: YAO Matrix <matrix.yao@intel.com>
BenjaminBossan
left a comment
There was a problem hiding this comment.
Generally, the PR LGTM. I tested running an experiment, the logs look correct and the app is working as expected. I wonder if we can find better names than "accelerator_memory_reserved_avg", as "accelerator" is quite a mouthful, but I can live with it.
@yao-matrix can you confirm that the script runs successfully on XPU? Could you perhaps run a LoRA experiment and share the result log in a comment?
method_comparison/app.py
Outdated
| with gr.Row(): | ||
| x_default = ( | ||
| "cuda_memory_max" if "cuda_memory_max" in metric_preferences else list(metric_preferences.keys())[0] | ||
| "accelerator_memory_max" if "accelerator_memory_max" in metric_preferences else list(metric_preferences.keys())[0] |
There was a problem hiding this comment.
Could you please add line breaks (or run ruff format app.py)?
There was a problem hiding this comment.
done with ruff format, thx.
|
@yao-matrix Could you please merge with/rebase on the latest main again? |
yes, the script runs successfully on XPU. I lost my dev env during weekends(since it's shared machine), pls allow me some time to find another env and get the result log. |
seems done. |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
@yao-matrix it seems like the merge was not with the latest version of PEFT main, could you please fetch it and try again? |
|
done, thx @BenjaminBossan |
|
failing tests are caused by latest transformers release, please ignore |
|
The issue should be fixed, could you please merge/rebase again @yao-matrix 🙏 |
|
done, thx, @BenjaminBossan |
BenjaminBossan
left a comment
There was a problem hiding this comment.
Sorry for the delay. The PR LGTM, thanks.
@githubnemo I think it's good to merge, anything you'd like to add?
|
ping @githubnemo |
…lerators like XPU (huggingface#2610) make method comparision device agnostic, so it can expand to more accelerators like XPU --------- Signed-off-by: YAO Matrix <matrix.yao@intel.com>
@BenjaminBossan @githubnemo , #2606 corrupted while rebasing, so I opened another PR here. Pls help review, thx very much.