database-fork/Poro.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: Finland
# - Contributor type: Academic (Research community)

system:
    name: Poro
    link: https://huggingface.co/LumiOpen/Poro-34B
    type: text
    performanceclass: full
    basemodelname: Poro-34B
    endmodelname: Poro-34B
    endmodellicense: Apache-2.0
    releasedate: 2024-04
    notes: Bilingual model trained on Finnish, English, and programming languages

org:
    name: Silo AI and TurkuNLP and High Performance Language Technologies (HPLT)
    link: ["https://www.silo.ai", "https://turkunlp.org", "https://hplt-project.org"]
    notes: Silo AI was acquired by AMD in August 2024

# availability:
datasources_basemodel:
    class: partial
    link: https://arxiv.org/html/2404.01856v1
    notes: Even though all datasets are listed in the pre-print, some resources are no longer available

datasources_endmodel:
    class: partial
    link: https://arxiv.org/html/2404.01856v1
    notes: Same as datasources_basemodel; no additional fine-tuning data specified.

weights_basemodel:
    class: open
    link: https://huggingface.co/LumiOpen/Poro-34B
    notes: Model weights available at various training checkpoints.

weights_endmodel:
    class: open
    link: https://huggingface.co/LumiOpen/Poro-34B
    notes: Final model weights released under Apache 2.0 license.

trainingcode:
    class: open
    link: https://github.com/LumiOpen/Megatron-DeepSpeed
    notes: Custom fork of the Megatron-Deepspeed framework used for training Poro-34B.

# documentation:
code:
    class: open
    link: https://github.com/TurkuNLP/Megatron-DeepSpeed/tree/main
    notes: Training code and related scripts are publicly available.

hardware_architecture:
    class: open
    link: https://arxiv.org/html/2404.01856v1
    notes: Hardware architecture described in paper.

preprint:
    class: open
    link: https://arxiv.org/html/2404.01856v1
    notes: Preprint available on arXiv.

paper:
    class: open
    link: https://dspace.ut.ee/items/4556afe0-53bb-499d-a012-fa72a63cdd4d
    notes: Paper published in the proceedings of NoDaLiDa/Baltic-HLT 2025

modelcard:
    class: open
    link: https://huggingface.co/LumiOpen/Poro-34B
    notes: Model card provides broad overview and links to full details.

datasheet:
    class: open
    link: [https://huggingface.co/datasets/LumiOpen/Poro-34B-dataset,https://arxiv.org/html/2404.01856v1 ]
    notes: All details provided on HuggingFace and pre-print

# access:
package:
    class: closed
    link:
    notes: No Packages published

api:
    class: closed
    link:
    notes: No API found
    metaprompts: closed

licenses:
    class: open
    link: https://huggingface.co/LumiOpen/Poro-34B#license
    notes: Apache 2.0 license, an OSI-approved license