Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions CONTRIBUTING.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,13 @@
# Contributing To Trace

## Community

Use [GitHub Issues](https://github.com/maveryn/trace/issues) for concrete,
scoped work and reproducible bug reports. Use
[GitHub Discussions](https://github.com/maveryn/trace/discussions) for
questions, integration ideas, and reproduction reports that are not yet a
specific change request.

## Development Setup

```bash
Expand Down
25 changes: 19 additions & 6 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,14 +7,15 @@ post-training. Each example includes an image, prompt, typed answer,
image-space annotation, verifier metadata, and execution trace. The registry
contains 1,000 tasks across 277 scenes and 11 visual domains.

Training Qwen2.5-VL-3B and 7B on 64,000 Trace instances improved the
macro-average across 24 external benchmarks by **+3.51** and **+4.06 percentage
points**, respectively. [See the transfer results](#external-benchmark-transfer).

<p align="center">
<a href="https://maveryn.github.io/trace/"><img alt="Project Page" src="https://img.shields.io/badge/Project-Page-14324a?style=for-the-badge"></a>
<a href="https://huggingface.co/spaces/maveryn/trace"><img alt="Live Demo" src="https://img.shields.io/badge/HuggingFace-Live_Demo-2563eb?style=for-the-badge"></a>
<a href="https://arxiv.org/abs/2607.19790"><img alt="arXiv" src="https://img.shields.io/badge/arXiv-2607.19790-b31b1b?style=for-the-badge"></a>
<a href="https://huggingface.co/collections/maveryn/trace-6a604291b4be4ed6399b9f24"><img alt="Hugging Face Collection" src="https://img.shields.io/badge/HuggingFace-Collection-f59e0b?style=for-the-badge"></a>
<a href="https://arxiv.org/abs/2607.19790"><img alt="Paper" src="https://img.shields.io/badge/arXiv-Paper-b31b1b?style=for-the-badge"></a>
<a href="https://huggingface.co/datasets/maveryn/trace"><img alt="Dataset" src="https://img.shields.io/badge/HuggingFace-Dataset-f59e0b?style=for-the-badge"></a>
<a href="https://huggingface.co/maveryn/trace-qwen2.5-vl-3b"><img alt="3B Model" src="https://img.shields.io/badge/HuggingFace-3B_Model-f59e0b?style=for-the-badge"></a>
<a href="https://huggingface.co/maveryn/trace-qwen2.5-vl-7b"><img alt="7B Model" src="https://img.shields.io/badge/HuggingFace-7B_Model-f59e0b?style=for-the-badge"></a>
<a href="https://github.com/maveryn/trace/blob/main/docs/research/REPRODUCING_RESULTS.md"><img alt="Reproduce" src="https://img.shields.io/badge/Results-Reproduce-159b91?style=for-the-badge"></a>
</p>

![Trace examples across charts, games, geometry, graphs, icons, illustrations, pages, physics, puzzles, symbolic reasoning, and 3D scenes](docs/assets/paper-domain-montage/trace-paper-domain-montage.png)
Expand Down Expand Up @@ -101,6 +102,9 @@ The table reports all 24 external benchmarks for the matched 3B and 7B
comparisons. Parenthesized values are mean seed-paired TRACE-minus-Base changes
in percentage points.

<details>
<summary><strong>View all 24 benchmark results</strong></summary>

<!-- trace-eval-v1-base-trace-table:start -->
| Benchmark | 3B Base | 3B TRACE (Δ) | 7B Base | 7B TRACE (Δ) |
| --- | ---: | ---: | ---: | ---: |
Expand Down Expand Up @@ -136,12 +140,18 @@ in percentage points.
| MME-Reasoning | 22.62 ± 1.81 | 25.06 ± 1.53 (+2.44) | 25.17 ± 1.18 | 28.03 ± 0.89 (+2.86) |
<!-- trace-eval-v1-base-trace-table:end -->

</details>

The [research page](docs/research/README.md#external-benchmark-results) includes
the full eight-model comparison. Machine-readable per-seed scores are available
in [`results.json`](https://github.com/maveryn/trace/blob/rlvr/rlvr/evaluation/trace_eval/results.json).

## Released Models

The [Trace collection](https://huggingface.co/collections/maveryn/trace-6a604291b4be4ed6399b9f24)
groups the interactive Space, dataset, checkpoints, paper, and evaluation
artifacts in one place.

| Released checkpoint | Base model |
| --- | --- |
| [TRACE Qwen2.5-VL-3B](https://huggingface.co/maveryn/trace-qwen2.5-vl-3b) | [`Qwen2.5-VL-3B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct) |
Expand Down Expand Up @@ -344,7 +354,10 @@ attributions are listed in
## Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, testing, review, and pull
request guidance.
request guidance. Browse [open issues](https://github.com/maveryn/trace/issues)
for scoped work, or use
[Discussions](https://github.com/maveryn/trace/discussions) for questions,
integration ideas, and reproduction reports.

## Licensing

Expand Down
8 changes: 8 additions & 0 deletions docs/DEVELOPMENT.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,13 @@
# Development

## Community

Use [GitHub Issues](https://github.com/maveryn/trace/issues) for concrete,
scoped work and reproducible bug reports. Use
[GitHub Discussions](https://github.com/maveryn/trace/discussions) for
questions, integration ideas, and reproduction reports that are not yet a
specific change request.

## Source Layout

- `src/trace_tasks/core/`: generation, identity, validation, trace, reward, and export infrastructure
Expand Down
21 changes: 17 additions & 4 deletions docs/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -26,15 +26,19 @@ hide:
trace.
</p>

<p class="trace-hero__evidence">
Training Qwen2.5-VL-3B and 7B on 64,000 Trace instances improved the
macro-average across 24 external benchmarks by <strong>+3.51</strong> and
<strong>+4.06 percentage points</strong>, respectively.
<a href="#external-benchmark-transfer">See the transfer results.</a>
</p>

<div class="trace-resource-links">
<a class="md-button md-button--primary" href="https://huggingface.co/spaces/maveryn/trace">Live Demo</a>
<a class="md-button" href="https://github.com/maveryn/trace">Code</a>
<a class="md-button" href="https://arxiv.org/abs/2607.19790">Paper</a>
<a class="md-button" href="research/">Research</a>
<a class="md-button" href="https://huggingface.co/collections/maveryn/trace-6a604291b4be4ed6399b9f24">Collection</a>
<a class="md-button" href="https://huggingface.co/datasets/maveryn/trace">Dataset</a>
<a class="md-button" href="https://huggingface.co/maveryn/trace-qwen2.5-vl-3b">3B Model</a>
<a class="md-button" href="https://huggingface.co/maveryn/trace-qwen2.5-vl-7b">7B Model</a>
<a class="md-button" href="research/REPRODUCING_RESULTS/">Reproduce</a>
</div>
</div>

Expand Down Expand Up @@ -201,6 +205,9 @@ The table reports all 24 external benchmarks for the matched 3B and 7B
comparisons. Parenthesized values are mean seed-paired TRACE-minus-Base changes
in percentage points.

<details markdown="1">
<summary><strong>View all 24 benchmark results</strong></summary>

<!-- trace-eval-v1-base-trace-table:start -->
| Benchmark | 3B Base | 3B TRACE (Δ) | 7B Base | 7B TRACE (Δ) |
| --- | ---: | ---: | ---: | ---: |
Expand Down Expand Up @@ -236,11 +243,17 @@ in percentage points.
| MME-Reasoning | 22.62 ± 1.81 | 25.06 ± 1.53 (+2.44) | 25.17 ± 1.18 | 28.03 ± 0.89 (+2.86) |
<!-- trace-eval-v1-base-trace-table:end -->

</details>

The [research page](research/README.md#external-benchmark-results) includes the
full eight-model comparison and links to the evaluation artifacts.

## Released models

The [Trace collection](https://huggingface.co/collections/maveryn/trace-6a604291b4be4ed6399b9f24)
groups the interactive Space, dataset, checkpoints, paper, and evaluation
artifacts in one place.

| Released checkpoint | Base model |
| --- | --- |
| [TRACE Qwen2.5-VL-3B](https://huggingface.co/maveryn/trace-qwen2.5-vl-3b) | [Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct) |
Expand Down
3 changes: 3 additions & 0 deletions docs/assets/brand/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,9 @@ scene families.
documentation headers.
- `trace-mark-light.png`: light-theme raster used by static documentation
composites.
- `../social/trace-social-preview.png`: the `1280x640` solid-background
repository and website social preview, generated by
`scripts/generate_social_preview.py`.

The SVG files are transparent and self-contained. Their structural ink adapts
to the viewer's light or dark color preference. The standalone mark has no font
Expand Down
Binary file added docs/assets/social/trace-social-preview.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
10 changes: 5 additions & 5 deletions docs/overrides/main.html
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@
{% set social_title = config.site_name %}
{% endif %}
{% set social_description = page.meta.description if page and page.meta and page.meta.description else config.site_description %}
{% set social_image = "https://maveryn.github.io/trace/assets/paper-domain-montage/trace-paper-domain-montage.png" %}
{% set social_image = "https://maveryn.github.io/trace/assets/social/trace-social-preview.png" %}
<meta property="og:type" content="website">
<meta property="og:site_name" content="{{ config.site_name | e }}">
<meta property="og:title" content="{{ social_title | e }}">
Expand All @@ -19,12 +19,12 @@
<meta property="og:url" content="{{ page.canonical_url | e }}">
{% endif %}
<meta property="og:image" content="{{ social_image }}">
<meta property="og:image:width" content="1512">
<meta property="og:image:height" content="780">
<meta property="og:image:alt" content="Examples from the eleven Trace visual domains and the Trace mark">
<meta property="og:image:width" content="1280">
<meta property="og:image:height" content="640">
<meta property="og:image:alt" content="Trace visual reasoning tasks and external benchmark transfer results">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="{{ social_title | e }}">
<meta name="twitter:description" content="{{ social_description | e }}">
<meta name="twitter:image" content="{{ social_image }}">
<meta name="twitter:image:alt" content="Examples from the eleven Trace visual domains and the Trace mark">
<meta name="twitter:image:alt" content="Trace visual reasoning tasks and external benchmark transfer results">
{% endblock %}
12 changes: 12 additions & 0 deletions docs/stylesheets/extra.css
Original file line number Diff line number Diff line change
Expand Up @@ -115,6 +115,18 @@ pre {
max-width: 48rem;
}

.trace-hero__evidence {
color: var(--trace-muted);
font-size: 0.84rem;
line-height: 1.6;
margin: 1.1rem auto 0;
max-width: 46rem;
}

.trace-hero__evidence strong {
color: var(--trace-teal-text);
}

.trace-resource-links {
align-items: center;
display: flex;
Expand Down
3 changes: 3 additions & 0 deletions scripts/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,9 @@ The tools are:
- `generate_release_gallery.py` regenerates the deterministic 11-domain
gallery under `docs/assets/gallery/`, its hero montage, and the manifest at
`docs/gallery/manifest.v1.json`.
- `generate_social_preview.py` composes the `1280x640` repository and website
social preview from the committed Trace mark, domain montage, and verified
headline results.

Scripts added here must use the `trace_tasks` namespace, accept
repository-relative or user-supplied paths, avoid credentials and uploads, and
Expand Down
6 changes: 3 additions & 3 deletions scripts/check_docs_site.py
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,10 @@
SITE_URL = "https://maveryn.github.io/trace/"
SITE_PATH_PREFIX = "/trace"
SOCIAL_IMAGE_URL = (
f"{SITE_URL}assets/paper-domain-montage/trace-paper-domain-montage.png"
f"{SITE_URL}assets/social/trace-social-preview.png"
)
SOCIAL_IMAGE_PATH = Path("assets/paper-domain-montage/trace-paper-domain-montage.png")
SOCIAL_IMAGE_SIZE = (1512, 780)
SOCIAL_IMAGE_PATH = Path("assets/social/trace-social-preview.png")
SOCIAL_IMAGE_SIZE = (1280, 640)
REQUIRED_PAGES = {Path("index.html"), Path("research/index.html")}
NONCANONICAL_PAGES = {Path("404.html")}
_SITE_PARTS = urlsplit(SITE_URL)
Expand Down
Loading
Loading