ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[ray] allow for specifying ray.init kwargs (i.e. runtime_env) (#7647)

Browse Source 
* ray init kwargs

* Update trainer_utils.py

* fix ray args

---------

Co-authored-by: hoshi-hiyouga <hiyouga@buaa.edu.cn>
This commit is contained in:
Eric Tang
2025-04-09 20:31:05 -07:00
committed by GitHub
parent 1c436c9f25
commit bb8d79bae2
3 changed files with 16 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
examples/train_lora/llama3_lora_sft_ray.yaml
View File

@@ -31,10 +31,16 @@ report_to: none # choices: [none, wandb, tensorboard, swanlab, mlflow]
### ray
ray_run_name: llama3_8b_sft_lora
ray_storage_path: ./saves
ray_num_workers: 4 # number of GPUs to use
ray_num_workers: 4 # Number of GPUs to use.
placement_strategy: PACK
resources_per_worker:
GPU: 1
placement_strategy: PACK
# ray_init_kwargs:
# runtime_env:
# env_vars:
# <YOUR-ENV-VAR-HERE>: "<YOUR-ENV-VAR-HERE>"
# pip:
# - emoji
### train
per_device_train_batch_size: 1