Announcing MOSS-TTS-PNY 1.7B v0.1
Finetune of the MOSS-TTS-Local model with fixed speaker embedding + a custom iSTFTNet3 vocoder that intercepts the MOSS tokenizer's features and outputs 48KHz audio.
Runs 1.8x realtime on a single RTX 5090 w/PyTorch:
(see replies)
>"post-trained to reinforce Western values"
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
1
141
Most devs I know: Yeah, I use Claude for everything, this is awesome
Lilith☭🏳️⚧️🇵🇸 (she/her): AI coding is theft, and it's buggy, insecure and evil because it steals water and jobs. Oh and Venmo me $20 for a burrito.
i don’t care if the ai is doing the “boring” parts of coding, people have been making video games just fine for decades before generative ai. if you’re facing a problem solving it yourself breeds these beautiful things called “experience” and “ingenuity”
2
20
662












