Konstantin PRO

apsys

whitecircle

·

AI & ML interests

Interpretable ML, classical ML, LLM & inference opts

Recent Activity

reacted to medmekk's post with 🚀 5 days ago

🚀 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

liked a model 10 days ago

whitecircle/GLM-4.7-Flash-Coder

posted an update 17 days ago

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

View all activity

Organizations

published an article 9 months ago

Article

TFLOPS Gap: Why FP4 MoE Kernel Engineering Matters on Blackwell

apsys

•

Jan 5

• 14

published an article over 1 year ago

Article

CircleGuardBench: New Standard for Evaluating AI Moderation Models

whitecircle

•

May 7, 2025

• 59