How to set the tune.yaml, to find the best lr at 4 nvidia gpu devices? - Githubissues

baudm / parseq

Scene Text Recognition with Permuted Autoregressive Sequence Models (ECCV 2022)

https://huggingface.co/spaces/baudm/PARSeq-OCR

Apache License 2.0

565 stars 126 forks source link

How to set the tune.yaml, to find the best lr at 4 nvidia gpu devices? #123

Closed Yullj closed 9 months ago

Yullj commented 10 months ago

How to set the tune.yaml, to find the best lr at 4 nvidia gpu devices?

`defaults:

main
self

trainer: gpus: 1 # tuning with DDP is not yet supported.

tune: num_samples: 10 gpus_per_trial: 1 lr: min: 1e-4 max: 2e-3 resume_dir: null

hydra: run: dir: rayresults/${model.name}/${now:%Y-%m-%d}${now:%H-%M-%S}`

baudm commented 9 months ago

Set CUDA_VISIBLE_DEVICES=0,1,2,3 when you run tune.py. It would run four trials in parallel.