github-featured/microsoft--agent-lightning

Files

T

History

Yuge Zhang a0791e8b13 Fix flaky pipeline (#131 )

2025-10-09 14:54:33 +08:00

..

Fix pyright issues (#101 )

2025-09-20 14:51:56 +08:00

README.md

Remove evaluation results (#6 )

2025-07-17 16:23:08 +08:00

sql_agent.py

fix: add BaseMessage import to sql_agent.py (#126 )

2025-10-09 01:39:41 +08:00

train_ci.sh

Fix flaky pipeline (#131 )

2025-10-09 14:54:33 +08:00

train_llama.sh

Update pre-commit configurations and bump to 0.2.0 (#102 )

2025-09-20 07:20:14 +00:00

train.sh

Update pre-commit configurations and bump to 0.2.0 (#102 )

2025-09-20 07:20:14 +00:00

README.md

Spider Example

This example requires a single node with one GPU of at least 40GB memory.

Download Spider 1.0 dataset from here and unzip it to the data folder.
Use python spider_eval/convert_dataset.py to convert the dataset to the parquet format.
Start ray: bash ../../scripts/restart_ray.sh. To use Wandb, you need to set the WANDB_API_KEY environment variable before starting ray.
Run the agent: VERL_API_BASE=http://localhost:9999/ python sql_agent.py. Use python sql_agent.py --help to see options like running multiple agents.
In another terminal, launch the training server: bash train.sh.

Evaluation

Results are coming soon.