README.md

---
license: apache-2.0
language:
  - en
library_name: transformers
pipeline_tag: image-text-to-text
base_model: OpenGVLab/InternVL3-8B-hf
tags:
  - vision-language-model
  - vlm
  - reasoning
  - perception
  - rlvr
  - grpo
  - icml-2026
---

# VLM-CapCurriculum-InternVL3-8B-Staged

A vision-language model post-trained from **OpenGVLab/InternVL3-8B-hf** with the staged, capability-dimension curriculum from
*"From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models"*
(ICML 2026).

This release is the **Stage-3 step-186** checkpoint, which gave the largest gain over the merged baseline among the four backbones we tried.

> **TL;DR.** Visual perception — not reasoning length — is the dominant bottleneck for visual reasoning in VLMs. We fix this by post-training along a **capability axis** (perception → textual reasoning → visual reasoning) rather than mixing all data together.

| Resource | Link |
|---|---|
| 📄 Paper | <TODO_PAPER_URL> |
| 💻 Code | https://github.com/UCSC-VLAA/VLM-CapCurriculum |
| 🌐 Project page | https://ucsc-vlaa.github.io/VLM-CapCurriculum |
| 🤗 Collection (model + data + eval) | https://huggingface.co/collections/UCSC-VLAA/vlm-capcurriculum-from-seeing-to-thinking-icml-2026-6a07691f944148ccb2b183b8 |

## Headline numbers (extended benchmark suite, AVG over 10 benchmarks)

| Setting | Extended AVG |
|---|:---:|
| InternVL3-8B (base) | 29.69 |
| InternVL3-8B + Merged training | 41.94 |
| **InternVL3-8B + Staged (this model)** | **45.71** |

Δ over merged: **+3.77** — the largest staged-vs-merged gap among the four backbones. Most striking is WeMath (+9.90 over merged), evidence that decoupling perception and reasoning is especially impactful for weaker base models. See Appendix Table 9 of the paper for the full per-benchmark breakdown.

## How it was trained

Three RLVR stages with GRPO (on top of [EasyR1](https://github.com/hiyouga/EasyR1)):

1. **Stage 1 — visual perception** on `UCSC-VLAA/VLM-CapCurriculum-Perception` (synthesised + filtered DOCCI MCQs).
2. **Stage 2 — textual reasoning** on `UCSC-VLAA/VLM-CapCurriculum-TextReasoning` (ORZ-Math-13k).
3. **Stage 3 — visual reasoning** on `UCSC-VLAA/VLM-CapCurriculum-VisualReasoning` (CLEVR-Math + GeoQA170K + Math PUMA + DocVQA + ArxivQA mix). **Released checkpoint: step 186 of Stage 3.**

InternVL3 needs damped optimisation in Stage 2 to avoid entropy explosion (`lr=3e-7`, `kl=5e-2`, `clip=0.15`, `max_grad_norm=0.5`) — these are baked into the launch scripts.

All three stages share **one** system / format prompt — see [Inference](#inference) below.

Detailed launch scripts: [`training/examples/internvl3_8b/`](https://github.com/UCSC-VLAA/VLM-CapCurriculum/tree/main/training/examples/internvl3_8b) in the code repo.

## Inference

The model expects the unified system prompt that it was trained against:

```
You FIRST think about the reasoning process as an internal monologue and then
provide the final answer. The reasoning process MUST BE enclosed within
<think> </think> tags. The final answer MUST BE put in \boxed{}.
i.e. <think> reasoning here </think> \boxed{final answer here}
```

Quick start with LMDeploy (the InternVL family is served via LMDeploy in our setup):

```bash
lmdeploy serve api_server UCSC-VLAA/VLM-CapCurriculum-InternVL3-8B-Staged \
  --server-port 23342 --tp 4
```

For VLMEvalKit-style benchmark eval, plug it in via the `InternVL3_8B_Staged` alias defined in [`evaluation/configs/models.py`](https://github.com/UCSC-VLAA/VLM-CapCurriculum/blob/main/evaluation/configs/models.py).

## Intended use & limitations

Intended for research on vision-language reasoning, post-training methodology, and capability-dimension curriculum learning. Inherits the safety / bias profile of the underlying InternVL3-8B backbone; we have not added additional alignment fine-tuning. Not recommended for high-stakes deployments without further evaluation.

Trained at the 8B parameter scale with 4096-token max prompt length and a fixed group size of 5. Behaviour at much longer contexts or substantially different prompt formats has not been characterised.

## License & citation

Released under **Apache-2.0**, matching the upstream backbone. If you use this model, please cite:

```bibtex
@inproceedings{vlmcapcurriculum2026,
  title  = {From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models},
  author = {Juncheng Wu and Hardy Chen and Haoqin Tu and Xianfeng Tang and Freda Shi and Hui Liu and Hanqing Lu and Cihang Xie and Yuyin Zhou},
  booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
  year   = {2026}
}
```

## Acknowledgements

Built on top of [EasyR1](https://github.com/hiyouga/EasyR1), [VLMEvalKit](https://github.com/open-compass/VLMEvalKit), and the [InternVL](https://huggingface.co/OpenGVLab) family.