SGLang Team Open-Sources Miles v0.1: 744B Model RL on 64 GPUs at 263s/Step
aigclink · x · 2026-09-13
The team behind SGLang has open-sourced Miles v0.1, a production-ready post-training framework that turns training a company-specific model from a costly gamble into an engineering task — spin it up with Docker and swap in your own data and tasks.
Key capabilities
- Focused on agentic RL: models learn by trial and error in real environments, using tools, operating terminals, and completing multi-step tasks — aiming for a "worker" rather than a "chatbot."
- Scales to frontier models: the paper's case study runs GLM-5.2 (744B-A40B) on terminal-use coding tasks across 64 NVIDIA GB300 GPUs, with a median step time of 263 seconds over the first 30 steps.
- Reliability: engineered against mid-run crashes that waste huge compute; Modal calls it "consistently stable over long runs."
- One framework covers SFT, RL, distillation, and LoRA, and extends to diffusion models.
- Data stays local: weights and training logs remain on-premises, a must for finance, healthcare, and government clients.
- Supports DeepSeek, Qwen, GLM, Kimi, Gemma, and Nemotron, on both NVIDIA and AMD GPUs.
- Already used in production by Periodic Labs (trillion-parameter training on thousands of GPUs), Modal, IBM, Decagon, and Nebius.
The author argues startups should seriously consider vertical models, enterprises now have controllable private-training costs, and AI service providers/FDEs may find a new market.
Related event: SGLang Team Open-Sources Miles v0.1 Production Post-Training Framework(2 posts)→
More from Infra
- Claymation body horror on 8GB: three wrong settings, not the GPU, made local renders slow — roro9- · 2026-10-30
- durable-object-migrator: zero-downtime exports for Cloudflare Durable Objects — samgoodwin89 · 2026-10-30
- An unnamed firm already runs a fairly scaled dedicated vehicle for compute lending, says analyst — matt_slotnick · 2026-10-30
- GPU break-even throughput vs open-weight token prices up 25-83% since May — bookwormengr · 2026-10-30
- Sam Altman says the chip supply chain's invisible hand is failing — and DRAM is why — SnoozeDoggyDog · 2026-10-30
- MoEShift runs a 125B MoE on a 24GB Mac: 7.5GB peak memory at ~10 tok/s — dave-tro · 2026-10-30