database-fork/Mistral.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: France
# - Contributor type: Non-academic (Company)

system:
    name: Mistral
    link: https://huggingface.co/mistralai/Mistral-Large-Instruct-2411
    type: text
    performanceclass: latest
    basemodelname: Mistral-Large-2411
    endmodelname: Mistral-Large-Instruct-2411
    endmodellicense: Mistral AI Research License
    releasedate: 2023-09
    notes: Dense LLM trained by Mistral AI.

org:
    name: Mistral AI
    link: https://mistral.ai/
    notes: Mistral, a French AI company.

# availability:
datasources_basemodel:
    class: closed
    link:
    notes: No information provided on pretraining data

datasources_endmodel:
    class: closed
    link: https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1
    notes: No information provided expect that instruction tuning is done using an unspecified 'variety of publicly available conversation datasets'

weights_basemodel:
    class: open
    link: https://github.com/mistralai/mistral-inference
    notes: Models available for download through their GitHub repository as .rar archives

weights_endmodel:
    class: open
    link: https://github.com/mistralai/mistral-inference
    notes: Models available for download through their GitHub repository as .rar archives

trainingcode:
    class: partial
    link: https://github.com/mistralai/mistral-src
    notes: repository provides 'minimal code to run our 7B model'

# documentation:
code:
    class: partial
    link: https://github.com/mistralai/mistral-inference/tree/main/src/mistral_inference
    notes: repository contains minimal code to run the models; also open source code, althought it is mostly uncommented and not documented very well.

hardware_architecture:
    class: partial
    link: https://github.com/mistralai/mistral-src
    notes: Some information on architecture provided in github repo

preprint:
    class: partial
    link: http://arxiv.org/abs/2310.06825
    notes: Preprint rehashes marketing blurbs also given in blog and provides no details about pretraining datasets, instruction tuning datasets, or fine-tuning process, hence partial.

paper:
    class: closed
    link:
    notes: No peer reviewed paper available

modelcard:
    class: closed
    link:
    notes: No model card available, HuggingFace modelcard just points to a corporate blog post

datasheet:
    class: closed
    link:
    notes: No datasheet available

# access:
package:
    class: open
    link: https://ollama.com/library/mistral-large
    notes: Model available on Ollama.

api:
    class: open
    link: https://docs.mistral.ai/api
    notes: API specification provided by vLLM
    metaprompt: closed

licenses:
    class: open
    link: https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/blob/main/README.md
    notes: Apache 2.0