Published at 11/14/2026, 1:20:00 AM · 64 views
📰 Thinking Machines Details End-to-End 4-Bit Training for Inkling 2 Small: NVFP4 Forward Passes in Expert Layers From Pretraining Through RL, 8% Faster…
In a Connectionism blog post, Thinking Machines explains how Inkling 2 Small used a native 4-bit (NVFP4) forward pass from the first pretraining step through the final RL rollout, with MXFP8 backward passes using stochastic rounding and quantized communications.
💬 What's your take? Join the discussion in the comments.
Read the original Chinese version
📰 Thinking Machines 公开 Inkling 2 Small 端到端 4 位训练方案：专家层前向全程 NVFP4，预训练提速 8%、RL 提速 20%，594B 模型 68 小时完成预训练
Thinking Machines 在 Connectionism 博客发文，介绍 Inkling 2 Small 如何从预训练第一步到最后一次 RL rollout 都用原生 4 位（NVFP4）前向计算：反向用 MXFP8 加随机舍入，通信也做量化。
💬 你怎么看？欢迎在评论区讨论。
#Thinking Machines #Inkling 2 Small #NVFP4
0 Comments