GeoWeave¶
Product or Programme · All products and programmes
Evidence · 2 claims (2 for) · 1 study (1 design),q2· 0 of 1 report an effect size · 2 claims rest on one study
Description¶
A post-training framework developed by Zhitong Dong et al. that combines capability-ordered supervised fine-tuning, interleaved reinforcement learning, and step-level credit assignment to optimize multimodal geometric reasoning models.
Components¶
Claims¶
- GeoWeave-8B achieves the highest open-source final-answer accuracy and process average on GeoVAD-Bench, gaining 25.3 and 30.4 percentage points over its base model [+M]
- Step-level credit assignment added to Interleave-RL raises the reasoning score from 46.3% to 66.3% on GeoVAD-Bench [+M]
Related Products and Programmes¶
-
Key Sources¶
- Zhitong Dong, Jicai Pan, Yingguo Gao, Jingting Ding, Hao Chen, Jinjie Gu. (2026). Beyond Generation and Accuracy: Diagnosing and Enhancing Visual Chain-of-Thought for Geometry Problem Solving. arXiv preprint. https://arxiv.org/abs/2609.12606
Page metadata
| Field | Value |
|---|---|
type |
product |
id |
geoweave |
title |
GeoWeave |
description |
A post-training framework developed by Zhitong Dong et al. |
status |
review |
product_kind |
framework |
generated |
by process:settle-candidates on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
Frontmatter is the page's machine-readable record — see the Schema & Guide.