File size: 5,543 Bytes
8b54ccf
078f872
 
 
 
 
 
8724a30
 
 
 
 
078f872
 
 
8724a30
078f872
8724a30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
078f872
8724a30
 
078f872
8724a30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1608675
8724a30
 
 
 
 
1608675
8724a30
 
 
 
078f872
8b54ccf
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Project Gaia</title>
<style>
  .gaia { font-family: system-ui, -apple-system, "Segoe UI", sans-serif; max-width: 760px; margin: 0 auto; padding: 8px 0; line-height: 1.55; }
  .gaia h2 { margin-top: 1.4em; }
  .gaia table { border-collapse: collapse; width: 100%; }
  .gaia td, .gaia th { border-top: 1px solid rgba(127,127,127,0.3); padding: 6px 8px; vertical-align: top; text-align: left; }
  .gaia pre { white-space: pre-wrap; font-size: 0.85em; padding: 10px; border: 1px solid rgba(127,127,127,0.3); border-radius: 6px; }
</style>
</head>
<body>
<div class="gaia">
<h1>Project Gaia</h1>
<p><strong>An open research project: a small AI that knows UK pet-care rules properly — and knows when to say “ask your vet”.</strong></p>
<p>Project Gaia is a research project by MyPetSitter Ltd, Harrogate, UK. It asks whether a small, specialised language model, trained only on material it is allowed to use, can be more reliable on this one subject than a large general one — answering from the records it is given, citing them, carrying their cautions and vet referrals, and saying so when the records do not answer. It also asks whether a vocabulary and data mix trained jointly for English and Romanian gives Romanian a fair share of a model's capacity. Round 1 covers English and Romanian.</p>
<p>The models were trained from scratch on <strong>Isambard-AI</strong>, under UKRI's AI Research Resource (AIRR) award.</p>
<p>Website: <a href="https://projectgaia.co.uk"><strong>https://projectgaia.co.uk</strong></a></p>

<h2>Round 1 — released 12 November 2026</h2>
<table>
  <tr><th></th><th>release</th><th>what it is</th></tr>
  <tr><td>model</td><td><a href="https://huggingface.co/ProjectGaiaUK/gaia-3b-chat"><strong>gaia-3b-chat</strong></a></td><td>The Round 1 chat model, version 1.0: 3 billion parameters, English and Romanian, 131,072-token context, fine-tuned to answer pet-care questions from numbered records in its prompt, citing them, and to say when they do not answer. A research model: read its “Known limits and risks” first.</td></tr>
  <tr><td>model</td><td><a href="https://huggingface.co/ProjectGaiaUK/gaia-3b-base-128k"><strong>gaia-3b-base-128k</strong></a></td><td>The 3B base model it is made from: pre-trained from scratch on 150 billion tokens of openly licensed English, Romanian, pet-care, code and maths text, then extended to a 131,072-token context. It continues text; it is not chat-tuned.</td></tr>
  <tr><td>dataset</td><td><a href="https://huggingface.co/datasets/ProjectGaiaUK/gaia-records-bench"><strong>gaia-records-bench</strong></a></td><td>The Gaia Records Benchmark v0.1: held-out pet-care and animal-law questions in English and Romanian, answered from numbered official records; it measures citing the right record, carrying its cautions and vet referrals, and saying when the records do not answer.</td></tr>
</table>
<p>Each model card says how every figure was measured and who read it, and ships with the data card that names every source of training data and its licence.</p>

<h2>Licence</h2>
<ul>
  <li><strong>Models:</strong> Apache-2.0.</li>
  <li><strong>Benchmark:</strong> code and results Apache-2.0; each item keeps its source's licence — the Open Government Licence v3.0 (English sets) and CC BY 4.0 (Romanian set). The benchmark's <code>LICENSES.md</code> and <code>items/ATTRIBUTION.md</code> give the attribution.</li>
</ul>

<h2>How to cite</h2>
<p>The benchmark (from its <code>CITATION.cff</code>):</p>
<pre>@misc{petrisor2026gaiarecordsbench,
  author = {Petrisor, Marian V.},
  title  = {Gaia Records Benchmark: pet care and animal law, English--Romanian},
  year   = {2026},
  note   = {Version 0.1.0, released 2026-11-12},
  url    = {https://huggingface.co/datasets/ProjectGaiaUK/gaia-records-bench}
}</pre>
<p>The models:</p>
<pre>@misc{petrisor2026gaia3b,
  author = {Petrisor, Marian V.},
  title  = {Project Gaia Round 1: gaia-3b-chat and gaia-3b-base-128k},
  year   = {2026},
  note   = {Round 1, version 1.0, released 2026-11-12. Apache-2.0},
  url    = {https://huggingface.co/ProjectGaiaUK}
}</pre>
<p>And Isambard-AI, as its acknowledgement asks: McIntosh-Smith, S., Alam S. R. and Woods, C. (2024). “Isambard-AI: a leadership class supercomputer optimised specifically for Artificial Intelligence”. <a href="https://doi.org/10.48550/arXiv.2410.11199">https://doi.org/10.48550/arXiv.2410.11199</a></p>

<h2>Acknowledgement</h2>
<p>The authors acknowledge the use of resources provided by the Isambard-AI National AI Research Resource (AIRR). Isambard-AI is operated by the University of Bristol and is funded by the UK Government’s Department for Science, Innovation and Technology (DSIT) via UK Research and Innovation; and the Science and Technology Facilities Council [ST/AIRR/I-A-I/1023].</p>
<p>UKRI AI Research Resource (AIRR) Gateway · project 0261-1548-0489-1 · 4,000 GPU hours of computing time.</p>
<p>The award supports the research. It is not an endorsement of any product.</p>
<p>Cite: McIntosh-Smith, S., Alam S. R. and Woods, C. (2024). “Isambard-AI: a leadership class supercomputer optimised specifically for Artificial Intelligence”. <a href="https://doi.org/10.48550/arXiv.2410.11199">https://doi.org/10.48550/arXiv.2410.11199</a></p>

<hr>
<p>Project Gaia · MyPetSitter Ltd, UK · <a href="https://projectgaia.co.uk">https://projectgaia.co.uk</a></p>
</div>
</body>
</html>