database-fork/Pythia.yaml at preview · Language-Technology-Assessment/database-fork · 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
98
99
100
101
102
103
104
---
# 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: International
# - Contributor type: Non-academic (Company)

system:
    name: Pythia
    link: https://huggingface.co/togethercomputer/Pythia-Chat-Base-7B
    type: text
    performanceclass: full
    basemodelname: Pythia-6.9B
    endmodelname: Pythia-Chat-Base-7B
    endmodellicense: Apache-2.0
    releasedate: 2023-03
    notes: Open-source model and open-source tool for building chatbots.

org:
    name: EleutherAI and Together Computer
    link: ["https://huggingface.co/EleutherAI", "https://together.ai/"]
    notes: Base model developed by EleutherAI, a major player in the open-source space. Fine-tune by Together Computer, a cloud platform for generative AI.

# availability:
datasources_basemodel:
    class: open
    link: https://huggingface.co/datasets/EleutherAI/the_pile_deduplicated
    notes: Training data published on HuggingFace.

datasources_endmodel:
    class: open
    link: https://github.com/togethercomputer/OpenDataHub
    notes: End model data shared on GitHub.

weights_basemodel:
    class: open
    link: https://huggingface.co/EleutherAI/pythia-6.9b
    notes: Model weights available via HuggingFace.

weights_endmodel:
    class: open
    link: https://huggingface.co/togethercomputer/Pythia-Chat-Base-7B
    notes: Model weights available via HuggingFace.

trainingcode:
    class: open
    link: https://github.com/togethercomputer/OpenChatKit/tree/main/training
    notes: Training code is made available through GitHub.

# documentation
code:
    class: open
    link: https://github.com/togethercomputer/OpenChatKit
    notes: Well-documented repository.

hardware_architecture:
    class: open
    link: https://github.com/togethercomputer/OpenChatKit#reproducing-pythia-chat-base-7b
    notes: Architecture and full recipe for reproducing model provided.

preprint:
    class: partial
    link: https://arxiv.org/abs/2304.01373
    notes: Preprint describes LM base (Pythia) but not instruction tuning details.

paper:
    class: partial
    link: https://arxiv.org/abs/2304.01373
    notes: Foundational paper has been peer-reviewed.

modelcard:
    class: partial
    link: https://huggingface.co/togethercomputer/Pythia-Chat-Base-7B
    notes: Model card partially available but does not go into much detail about model specifics.

datasheet:
    class: partial
    link: ["https://huggingface.co/datasets/EleutherAI/the_pile_deduplicated", "https://github.com/togethercomputer/OpenDataHub"]
    notes: Only end model dataset documented.

# access
package:
    class: partial
    link: https://pypi.org/project/openchatkit/
    notes: Package is available on PyPi.

api:
    class: closed
    link:
    notes: Reference seems to be made to a feedback app on HuggingFace, however the link redirects to a 404 page.
    metaprompt: closed

licenses:
    class: open
    link: https://huggingface.co/togethercomputer/Pythia-Chat-Base-7B#model-details
    notes: Apache 2.0