Dev shares training dashboard: ~$11/b tokens cost with 'insane' MFU
jon_durbin · x · 2026-09-11
jondurbin shares a quick dashboard giving visibility into a training run: roughly $11 per billion tokens, which he calls insane, along with remarkable MFU. He describes the whole effort as "pretty legendary," notes the inference side, and mentions using a few nodes from @liumio.
Related event: Dev open-sources 1T-token pretraining dataset at ~$11 per billion tokens(4 posts)→
More from Infra
- Community challenge pits MLX vs CUDA to speed up local Qwen 3.8 Flash on DGX Spark — gajesh · 2026-09-11
- Straight-up FP4 KV cache with no degradation: inference engineers take note — stochasticchasm · 2026-09-11
- SGLang and Datawhale launch zero-to-sglang, the official AI infra course — calabi_and_yau · 2026-09-11
- New attention design called inference-informed: cheap prefill, small KV cache — stochasticchasm · 2026-09-11
- Post-training's research era ended months ago: data and environment engineering now beats algorithmic novelty — josh_wills · 2026-09-11
- NVIDIA open-sources BioNeMo Inference Runtime to speed up protein structure prediction — AllThingsApx · 2026-09-11