forked from trotsky1997/OpenFugu
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathNOTICE
More file actions
48 lines (36 loc) · 2.36 KB
/
Copy pathNOTICE
File metadata and controls
48 lines (36 loc) · 2.36 KB
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
OpenFugu
Copyright 2026 The OpenFugu Contributors
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this project except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
================================================================================
ATTRIBUTION & THIRD-PARTY MATERIAL
================================================================================
OpenFugu is an independent, open reverse-engineering and reimplementation of
the ideas behind Sakana AI's "Fugu" orchestrator family. It is NOT affiliated
with, endorsed by, or derived from Sakana AI's proprietary product code.
This repository contains ONLY original code, documentation, and analysis
authored by the OpenFugu contributors. It deliberately does NOT redistribute any
third-party source, weights, or datasets. Those are fetched at setup time by
`scripts/fetch_artifacts.py` directly from their original, licensed sources:
1. TRINITY: An Evolved LLM Coordinator (arXiv:2512.04695), Sakana AI.
- The released router checkpoint `model_iter_60.npy` and the 37-case eval
fixture are pulled from the third-party HuggingFace dataset
`nshkrdotcom/trinity-coordinator-adapted-qwen3-0.6b` (MIT) and the
authors' code submission. Not redistributed here.
2. Learning to Orchestrate Agents in Natural Language with the Conductor
(arXiv:2512.04388), Sakana AI. Referenced for the Conductor mechanism.
3. trinity_coordinator (github.com/nshkrdotcom/trinity_coordinator, MIT) —
source of the routing eval fixture used by `verify/verify_37.py`.
4. Backbone model: Qwen/Qwen3-0.6B (Apache-2.0), Alibaba Cloud.
5. Worker pool: any OpenAI-compatible / litellm-supported models. Credentials
are supplied by the user via environment variables and never stored.
================================================================================
TRAINED WEIGHTS
================================================================================
The Conductor checkpoint trained by this project (`openfugu-conductor-*`) is a
fine-tune of Meta's Llama-3.2-3B-Instruct and is therefore subject to the
LLAMA 3.2 COMMUNITY LICENSE AGREEMENT (https://llama.com/llama3_2/license/),
in addition to Apache-2.0 for the surrounding OpenFugu code. The weights are
distributed separately on HuggingFace, not in this git repository.