Downloads · 30 days
0
jtp-3-ONNX-akasha
inflatebot/jtp-3-ONNX-akasha
jtp-3-ONNX-akasha is a machine learning model from inflatebot. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
An ONNX-converted version of JTP-3, for use with Akasha, a desktop-native image gallery for data hoarders. Support is currently experimental.
Access
Public
Updated Jul 24, 2026
Repo size
5 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.data3 GB · 100%
At a glance
- License
- apache-2.0
- Access
- Public
- Created
- Jul 1, 2026
- Updated
- Jul 24, 2026
- SHA
- 032ca8c9
Base models
Model card
From the Hugging Face model README
NOTE: JTP-3 has been replaced with Hydra-3.5.
Rather than re-convert the model, we're writing a custom Rust implementation for Hydra, as ONNX performed poorly for JTP-3 and GPU acceleration was inaccessible.
This repo will be kept up for posterity!
An ONNX-converted version of JTP-3, for use with Akasha, a desktop-native image gallery for data hoarders. Support is currently experimental.
Kimi K2.7 converted the weights using export_onnx_v2.py. All functions exposed in RedRocket's inference code, to my knowledge, is ported to Akasha. The only thing we don't currently have is extensions; this is pending a potential rewrite of ONNX handling in general (considering a move to Burn) as it's the most involved part by far.
To use, download jtp-3-hydra-fpXX.onnx and its matching .onnx.data, as well as tags.csv, tags.txt and pos_embed.safetensors, and aim an Akasha [[models]] table at it.
Currently JTP support is only available on Akasha's dev branch, and isn't currently compiled into nightly binaries.
More Documentation Soon:tm:
- License
- apache-2.0
- Created
- Jul 1, 2026
- Updated
- Jul 24, 2026