CircleGuardBench: New Standard for Evaluating AI Moderation Models
• 59
AI & ML interests
Runtime AI safety and alignment infrastructure
Articles
Team members 11 private
Post
4,392
🚀 Introducing Halo 1.0
Today, we are open-sourcing Halo, the training framework we use to train every model at White Circle.
It comes with:
🧠 Full post-training stack: SFT, DPO/KTO/SMPO, reward modeling, GRPO, distillation
🤖 Async multi-turn RL with vLLM/SGLang rollouts and sandboxed tool use
⚡ ~2.8× TRL throughput on 8× B300 (EP+FSDPv2, FA4, fp8/fp4)
🤗 Dense HF models + 15 MoE families (Qwen, GLM, Mistral, DeepSeek-V4…)
🛠️ One halo command, prebuilt Docker images, and docs for humans and agents
💻 https://github.com/whitecircle/halo
Try it and tell us what you're training
Today, we are open-sourcing Halo, the training framework we use to train every model at White Circle.
It comes with:
🧠 Full post-training stack: SFT, DPO/KTO/SMPO, reward modeling, GRPO, distillation
🤖 Async multi-turn RL with vLLM/SGLang rollouts and sandboxed tool use
⚡ ~2.8× TRL throughput on 8× B300 (EP+FSDPv2, FA4, fp8/fp4)
🤗 Dense HF models + 15 MoE families (Qwen, GLM, Mistral, DeepSeek-V4…)
🛠️ One halo command, prebuilt Docker images, and docs for humans and agents
💻 https://github.com/whitecircle/halo
Try it and tell us what you're training
- 1 reply
Post
90
43% less answer time in a looped transformer by changing answer supervision
check out https://x.com/advprop/status/2098087083470373010?s=20 + full writeup linked there
check out https://x.com/advprop/status/2098087083470373010?s=20 + full writeup linked there
apsys
authored a
paper over 2 years ago