Kxck
/

Finance_2408_v1_SFT

vision-language

Model card Files Files and versions

FinChart Phase 2D SFT-2408

This adapter continues Kxck/Finance_500_v1 on 2,408 ChartQA train-only records:

408 validated structured SFT examples from train[0:500];
2,000 disjoint answer-supervised examples from train[500:2500].

The frozen evaluation splits were excluded from training. This is an SFT scaling experiment, not a DPO model.

Frozen evaluation

ChartQA val[0:500]: 342 / 500 = 68.4%
ChartQA test[0:2500]: 1,903 / 2,500 = 76.12%

Inference used private vLLM, greedy decoding, and the Phase 1 exact-or-numeric-tolerance matcher. Full manifests are included in this repository.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Kxck/Finance_2408_v1_SFT

Base model

Qwen/Qwen3-VL-4B-Instruct

Finetuned

(445)

this model