Modelpublic

namhokaist/appgen-qwen3-thinktag-lr5e-7-ep1-continue-sft-amex-mobile-use

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face

0likes3downloads

Model Card

namhokaist/appgen-qwen3-thinktag-lr5e-7-ep1-continue-sft-amex-mobile-use

Continue-SFT of luca0621/appgen-qwen3-thinktag-lr5e-7-ep1 on AMEX-derived GUI trajectories converted to the mobile_use tool-call format used by AndroidWorld-style mobile agents.

—Base model: luca0621/appgen-qwen3-thinktag-lr5e-7-ep1
—Output format: <tool_call>{"name": "mobile_use", "arguments": ...}</tool_call>
—Coordinate format: absolute screen pixels
—Qwen3 target format: empty <think></think> block followed by the tool call
—Training data: 38,709 AMEX-derived mobile-use examples
—Training: 1 epoch, lr 5e-7, full SFT, 4x B200, per-device batch 1, gradient accumulation 4, max length 4096, max pixels 2629536
—W&B run: https://wandb.ai/namhokoh-korea-advanced-institute-of-science-and-technology/continuetrainappgenb200/runs/hs1idjly