AReno

[AReno @ARenoTeam]

AReno v0.0.8 is out.
This release brings native Apple Silicon support via MLX, PEFT-compatible LoRA on CUDA and MLX, and broader multimodal support for Gemma 4 and MiniCPM-V 4.6.
Try it here:
https://t.co/dDHYDhnCHU

[AReno @ARenoTeam]

AReno v0.0.7 is out.
New: Codebreaker TUI workflow, OLMo 2, Qwen3.5 image-to-text, Ling Tiny V3, rollout sync, and PPO/training CLI fixes.
Local RL post-training, faster:
https://t.co/2cznEQQOsJ
#LLM #RL #AgenticAI

[AReno @ARenoTeam]

Tic-tac-toe is just the smallest visible playground.
The same loop can extend to tool-call repair, structured extraction, workflow agents, and local task adaptation.
From running locally, to training locally.
Read more: [BLOG LINK]
GitHub: https://t.co/beiRFZRBMe

[AReno @ARenoTeam]

Can a model do more than run locally?
What if it could also learn locally?
We used AReno to post-train Ling-3.0-tiny on DGX Spark with Agentic RL.
The task: tic-tac-toe.
Small game. Full post-training loop. 🧵

[AReno @ARenoTeam]

The model also became more concise.
response_len dropped to ~850 tokens, meaning less wandering and more stable execution.
For agentic tasks, that matters: the model should not only do the right thing, but do it reliably.

[AReno @ARenoTeam]

Our first community meetup was a hit ✨

18 amazing folks showed up, the vibes were immaculate, and we had way more fun than a “meeting” usually promises.

Next one: Aug 21, 1 PM China Time.

Come join us. We’re accidentally becoming a movement.

[AReno @ARenoTeam]

Welcome to the AReno Community! 🎉
We hold biweekly Friday meetings to discuss releases, issues, features, and the future direction of AReno.
Everyone is welcome to join and contribute!
📅 Aug 7, Fri | 2:00–3:00 PM GMT+8
🔗 https://t.co/Bd0OhAf84y
Join via browser — no app needed

[AReno @ARenoTeam]

AReno v0.0.6 is out.

This release focuses on the practical developer workflow for RL post-training: setup, monitoring, packaging, and project navigation.

https://t.co/RkOoeUdyb4

[AReno @ARenoTeam]

AReno v0.0.4 is out: better SFT dataset support, GSM8K loader, clearer validation, rollout trainer tuning, PPO memory cleanup, stronger algorithm tests, and polished Docs V1.
Release note: https://t.co/QASucg3AiG

[AReno @ARenoTeam]

#新头像

[AReno @ARenoTeam]

AReno v0.0.3 is out 🎉

We shipped a few runnable agentic RL examples:

DuelGrid browser demo
Tic-Tac-Toe UI
Multi-turn coding training example
Plus some fixes for tool calls, CLI thinking toggle, and SFT/attention.

Release notes:
https://t.co/dkDZ3zO1Bh