Primary sourcerohanpaul_ai
View sourceEvent ·Multi-source
Thinking Machines Lab 开源 Inkling 2 Small:594B 参数 4-bit 前向训练模型
Thinking Machines Lab 兑现承诺,开源了 Inkling 2 Small,一个 594B 参数模型,训练时使用 4-bit 前向传播,已在 HuggingFace 上以 Apache 2.0 许可证提供下载。 4-bit 训练使预训练时间缩短 8%,RL 时间缩短 20%,效率损失约 2%。部署门槛不低:NVFP4 格式至少需要 500GB GPU 显存,MXFP8 则需要至少 800GB,而 MXFP8 在评测上仅比 NVFP4 高出 1-2 个百分点的边际收益。 技术上的关键是从预训练到最终 RL rollout 全程保持相同数值精度,因此发布的 4-bit checkpoint 无需单独的校准步骤。团队最初尝试前向和反向传播都用 NVFP4,但计算效率损失过大,最终只保留前向传播为 4-bit。
Current read
Thinking Machines 兑现了开源承诺,594B 参数的 Inkling 2 Small 直接上了 HuggingFace,Apache 20 许可可商用。4-bit 前向训练省了 8% 预训练和 20% RL 时间,只掉约 2% 效率,想研究低精度训练的团队值得拉下来跑跑。
10 sources84° AI heatLast updated 2026/11/13 23:32:05
Evidence
10 sourcesRelated source 1vLLM
View sourceRelated source 2lmarena.ai
View sourceRelated source 3Thinking Machines Lab
View sourceRelated source 4Gorden Sun
View sourceRelated source 5NVIDIA AI
View sourceRelated source 6Mira Murati (TML)
View sourceRelated source 7Together AI
View sourceRelated source 8ollama
View sourceRelated source 9TestingCatalog
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.