cua-lite/Lite.OSWorld-v1 at revision 0c5d2495f35a27eae7baf8aa50b2e677b209de39: 1,937 examples
cua-lite/Lite.CUAGym-v1 at revision 92fff449649d210bdfca09b71d4094900f8d5147: 1,788 examples
Combined total: 3,725 examples

The data was exported from examples/lite/v1/configs/qwen3_5/reasoning/lite.osworld.yaml, changing only agent_kwargs.resolution from [1024, 768] to [1920, 1080]. Qwen3.5's 32-pixel vision-grid alignment produces processed images of 1920x1088.

Training configuration

Epochs: 3
Global batch size: 32
Micro batch size: 1
Tensor parallel size: 4
Data parallel size: 2
Learning rate: 5e-6
Minimum learning rate: 1e-6
Final training step: 347
Final training loss: 0.219431

Only the final checkpoint was retained. The weights in this repository are the final Hugging Face export from Slurm job 1879517 (eval-bbq-2).

OnAnOrange/qwen3.5-9b-lite-osworld-cuagym-sft-1920x1080

Qwen3.5-9B Lite.OSWorld + Lite.CUAGym SFT (1920x1080)

Training data

Training configuration

OnAnOrange/
qwen3.5-9b-lite-osworld-cuagym-sft-1920x1080