main-database/CodeLlama.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: United States
# - Contributor type: Non-academic (American Big Tech)

# Training compute:
# - Base model training compute: ~8.1e+23 FLOP +- 0.5 OoM (geometric mean of param count & dataset site, and hardware figures) [Epoch AI]
# - End model training compute: ~4.2e+23 FLOP +- 0.5 OoM (param count & dataset size)

system:
    name: CodeLlama
    link: https://huggingface.co/meta-llama/CodeLlama-70b-Instruct-hf
    type: code
    performanceclass: full
    basemodelname: Llama-2-70B
    endmodelname: CodeLlama-70B-Instruct
    endmodellicense: Llama 2 Community License Agreement
    releasedate: 2024-03
    notes: Coder model by Meta.

org:
    name: Meta
    link: https://ai.meta.com/
    notes: Meta, a major technology company.

# availability:
datasources_basemodel:
    class: closed
    link: https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/
    notes: Data nowhere disclosed or documented, and described only in the vaguest terms in a corporate preprint released by Meta

datasources_endmodel:
    class: closed
    link:
    notes: Proprietary dataset used.

weights_basemodel:
    class: partial
    link: https://ai.meta.com/resources/models-and-libraries/llama-downloads/
    notes: Download only after requesting access; requires signing a consent form

weights_endmodel:
    class: partial
    link: https://huggingface.co/meta-llama/CodeLlama-70b-Instruct-hf
    notes: Gated model available on HuggingFace.

trainingcode:
    class: closed
    link: https://github.com/meta-llama/codellama
    notes: Repo exists, but does not contain training code.

# documentation:
code:
    class: closed
    link:
    notes: No training code, so undocumented.

hardware_architecture:
    class: partial
    link: "https://arxiv.org/pdf/2308.12950"
    notes: Hardware architecture discussed in aggregate and with a low level of detail in the model's paper.

preprint:
    class: open
    link: https://arxiv.org/pdf/2308.12950
    notes: Preprint available through arXiv.

paper:
    class: closed
    link: https://conf.researchr.org/details/icse-2024/llm4code-2024-papers/2/Code-Llama-Open-Foundation-Models-for-Code
    notes: No peer-reviewed paper found. Preprint presented as keynote in ICSE24.

modelcard:
    class: partial
    link: https://huggingface.co/meta-llama/CodeLlama-70b-Instruct-hf
    notes: Model card provides some information about training and inference, however mostly contains usage instructions.

datasheet:
    class: closed
    link:
    notes: No datasheet found.

# access:
licenses:
    class: closed
    link: https://huggingface.co/meta-llama/CodeLlama-70b-Instruct-hf#model-details
    notes: "Weights: Llama 2 Community License Agreement (not OSI-approved). Code: not released. Data: not disclosed."