shubhamgoel27

curbcheck-qwen25vl3b-v5-lora

README

License: mit

curbcheck v5 — QLoRA adapter for Qwen2.5-VL-3B (vision encoder unfrozen)

Reads San Francisco parking-sign stacks into structured rules. v5 trains on cross-city real photos (consensus-labeled) with the vision encoder unfrozen. Pair with the deterministic resolver in https://github.com/shubhamgoel27/curbcheck.

Base: Qwen/Qwen2.5-VL-3B-Instruct, QLoRA (language + vision layers)
Real SF photos: pipeline reasoning 0.90, end-to-end reasoning 0.82
Reads real sign-bearing photos at ~0.62 F1 (~0.52 vs a strict 3-vote consensus gold); abstains correctly on no-sign photos 83% of the time

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

shubhamgoel27

Model Tree

Base

Qwen/Qwen2.5-VL-3B-Instruct

Adapter

this model