main-database/Apertus.yaml at preview · Language-Technology-Assessment/main-database · GitHub

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
87
88
89
90
91
92
93
94
95
96
97
---
# Thank you for contributing!
# In filling out this yaml file, please follow the criteria as described here:
# https://osai-index.eu/contribute

# You're free to build on this work and reuse the data. It is licensed under CC-BY 4.0, with the
# stipulation that attribution should come in the form of a link to https://osai-index.eu/
# and a citation to the peer-reviewed paper in which the dataset & criteria were published:

# Liesenfeld, A. and Dingemanse, M., 2024. Rethinking open source generative AI: open-washing and the EU AI Act. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (pp. 1774-1787).

# Organization tags:
# - National origin: Switzerland
# - Contributor type: Academic (Research institution)

# Training compute:
# - Base model training compute: 6.74e+24 FLOP (from preprint)
# - End model training compute: unknown (likely negligible)

system:
    name: Apertus
    link: https://huggingface.co/collections/swiss-ai/apertus-llm-68b699e65415c231ace3b059
    type: text
    performanceclass: full
    basemodelname: Apertus-70B-2509
    endmodelname: Apertus-70B-Instruct-2509
    endmodellicense: Apache-2.0
    releasedate: 2025-09
    notes: Open-data open-weights models, multilingual in >1000 languages, developed at Swiss public universities.

org:
    name: Swiss AI Initiative
    link: https://swiss-ai.org
    notes: Swiss National AI Institute (non-profit research institute) collaboration with Swiss National Supercomputing Centre (CSCS).

# availability:
datasources_basemodel:
    class: open
    link: ["https://github.com/swiss-ai/pretrain-data", "https://huggingface.co/datasets/epfml/FineWeb2-HQ", "https://huggingface.co/datasets/swiss-ai/fineweb-2-compliant-tag", "https://huggingface.co/datasets/swiss-ai/fineweb-compliant-tag"]
    notes: Training data for base model released and use documented.

datasources_endmodel:
    class: open
    link: ["https://huggingface.co/swiss-ai/datasets"]
    notes: Data for fine-tuning published in a well-organized manner.

weights_basemodel:
    class: open
    link: https://huggingface.co/swiss-ai/Apertus-70B-2509/tree/main
    notes: Model weights available in many stages.

weights_endmodel:
    class: open
    link: https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/tree/main
    notes: Model weights available in many stages.

trainingcode:
    class: open
    link: https://github.com/swiss-ai/
    notes: Multiple repos with training, architecture and fine-tuning code available.

# documentation:
code:
    class: open
    link: https://github.com/swiss-ai/
    notes: Repositories and code well-described, commented and documented.

hardware_architecture:
    class: open
    link: https://github.com/swiss-ai/apertus-tech-report
    notes: Architecture documented in requisite detail.

preprint:
    class: open
    link: https://arxiv.org/abs/2509.14233
    notes: Preprint version of the paper available here.

paper:
    class: closed
    link:
    notes: Paper under peer review

modelcard:
    class: open
    link: https://huggingface.co/swiss-ai/Apertus-70B-2509/
    notes: Model card provides broad overview and links to full details.

datasheet:
    class: open
    link: ["https://huggingface.co/datasets/swiss-ai/fineweb-compliant-tag", "https://huggingface.co/datasets/swiss-ai/fineweb-2-compliant-tag", "https://huggingface.co/datasets/swiss-ai/fineweb-edu-compliant-tag", "https://huggingface.co/datasets/swiss-ai/robots-txt-blocked-domains-english", "https://huggingface.co/datasets/swiss-ai/robots-txt-blocked-domains-multilingual", "https://huggingface.co/datasets/swiss-ai/fineweb-robots-txt-files-compressed"]
    notes: Data sheets are well-documented and provide requisite info.

# access:
licenses:
    class: open
    link: ["https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/blob/main/LICENSE.txt", "https://github.com/swiss-ai/", "https://github.com/swiss-ai/pretrain-data", "https://huggingface.co/datasets/epfml/FineWeb2-HQ", "https://huggingface.co/datasets/swiss-ai/fineweb-2-compliant-tag", "https://huggingface.co/datasets/swiss-ai/fineweb-compliant-tag", "https://huggingface.co/swiss-ai/datasets"]
    notes: "Weights: Apache-2.0. Code: Apache-2.0. Data: OpenRAIL, ODC-By 1.0, ODC-BY."