Thinking Machines details end-to-end native 4-bit training of Inkling 2 Small on GB300
clarejtbirch · x · 2026-11-14
Thinking Machines Lab published a technical blog detailing how it trained Inkling 2 Small with a native 4-bit forward pass from the first pretraining step through the final RL rollout.
- The core challenge of low-precision training: fewer bits mean faster arithmetic, but rounding/quantization errors accumulate across layers and optimization steps and can impair learning.
- The team combines three numerical formats during training — BF16, MXFP8, and NVFP4. NVFP4 stores 3.6x fewer bits than BF16 and runs up to 6x faster on NVIDIA's GB300 (Grace Blackwell Ultra), at 15 PFLOPS vs 2.5 for BF16.
- The post explains how the tradeoff is managed to keep training stable while capturing NVFP4's speedup, pushing 4-bit acceleration from inference into full training.
Related event: Thinking Machines Details End-to-End 4-Bit Training of Inkling 2 Small(10 posts)→
More from Infra
- SemiAnalysis: Google TPUs match Nvidia in nightly benchmarks; the moat is Jensen's supply chain — elonmusk · 2026-11-14
- Musk warns Nvidia's handshake supply deals may unravel without Jensen — elonmusk · 2026-11-14
- Report: Nvidia veterans may bolt post-Jensen, most already financially free — basedjensen · 2026-11-14
- Jensen Huang calls TSMC chief directly over GPU batch delays, sparking org debate — IgorCarron · 2026-11-14
- Cisco logs strongest revenue growth in 15 years, credits agentic AI demand — Justgototheeffinmoon · 2026-11-14
- NVIDIA and TSMC reportedly never signed a formal contract in nearly 30 years of partnership — IgorCarron · 2026-11-14