File size: 4,391 Bytes
8b54ccf
078f872
 
 
 
8b54ccf
 
078f872
 
8b54ccf
 
078f872
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1608675
078f872
 
 
 
 
 
 
 
 
1608675
078f872
 
 
 
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
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
---
title: Project Gaia
emoji: 🌿
colorFrom: green
colorTo: blue
sdk: static
pinned: false
license: apache-2.0
short_description: Open research into an honest AI for UK pet care.
---

# Project Gaia

**An open research project: a small AI that knows UK pet-care rules properly — and knows when to say “ask your vet”.**

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.

The models were trained from scratch on **Isambard-AI**, under UKRI's AI Research Resource (AIRR) award.

Website: **https://projectgaia.co.uk**

## Round 1 — released 12 November 2026

| | release | what it is |
|---|---|---|
| model | [**gaia-3b-chat**](https://huggingface.co/ProjectGaiaUK/gaia-3b-chat) | 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. |
| model | [**gaia-3b-base-128k**](https://huggingface.co/ProjectGaiaUK/gaia-3b-base-128k) | 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. |
| dataset | [**gaia-records-bench**](https://huggingface.co/datasets/ProjectGaiaUK/gaia-records-bench) | 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. |

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.

## Licence

- **Models:** Apache-2.0.
- **Benchmark:** 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 `LICENSES.md` and `items/ATTRIBUTION.md` give the
  attribution.

## How to cite

The benchmark (from its `CITATION.cff`):

```bibtex
@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}
}
```

The models:

```bibtex
@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}
}
```

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”. https://doi.org/10.48550/arXiv.2410.11199

## Acknowledgement

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].

UKRI AI Research Resource (AIRR) Gateway · project 0261-1548-0489-1 · 4,000 GPU hours of computing time.

The award supports the research. It is not an endorsement of any product.

Cite: McIntosh-Smith, S., Alam S. R. and Woods, C. (2024). “Isambard-AI: a leadership class supercomputer optimised specifically for Artificial Intelligence”. https://doi.org/10.48550/arXiv.2410.11199

---

Project Gaia · MyPetSitter Ltd, UK · https://projectgaia.co.uk