portal-vlm: qwen25vl-lora-lm
Grounding LoRA (r16/a32), 252 LM-decoder sites, native absolute-pixel coordinate dialect. wave-ui-25k, 1 epoch, best checkpoint at step 372/1489 by held-out click probe.
Part of the portal-vlm release - an independent replication of Ramp Labs' PorTAL (portable task adapters via hypernet-generated LoRA) extended to vision-language models on GUI grounding.
| ScreenSpot-v2 overall | web split | |
|---|---|---|
| this artifact | 77.1% | 72.8% |
- Base model:
Qwen/Qwen2.5-VL-3B-Instruct@66285546d2b821cf421d4f5eb2576359d3770cd3 - Training config:
configs/qwen25vl_lora_lm.yaml - Eval record:
results/qwen25vl_lora_lm.json
Reproduce this row without training
git clone https://github.com/robbym-dev/portal-vlm && cd portal-vlm && uv sync
uv run python scripts/eval.py --config configs/qwen25vl_lora_lm.yaml --adapter hf:manihani4/portal-vlm-qwen25vl-lora-lm
Standard PEFT LoRA adapter - also loadable directly with peft.PeftModel.from_pretrained on the pinned base model.
- Downloads last month
- 10
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for manihani4/portal-vlm-qwen25vl-lora-lm
Base model
Qwen/Qwen2.5-VL-3B-Instruct