Event ·Multi-source

TML Inkling：1T参数开源多模态模型，支持百万token上下文

Thinking Machines Lab发布了TML Inkling，一个1T参数的开源权重模型，支持文本、图像和音频多模态输入。其上下文窗口高达100万token，采用相对注意力、短卷积和MoE专家汇聚的新架构，并配备8个MTP头用于推测解码。vLLM已从首日支持该模型，在NVIDIA Blackwell和Hopper上运行，4×GB200 MTP可达380 tok/s/user。

Current read

1T参数的开源多模态模型，百万token上下文，vLLM直接跑，4×GB200飙到380 tok/s，值得试试。

11 sources84° AI heatLast updated 2026/7/15 19:02:05

Evidence

11 sources

Primary sourcevLLM

2026/7/15 19:02:05View source

Related source 1The Rundown AI

2026/7/15 19:16:04View source

Related source 2marktechpost

2026/7/15 23:48:58View source

Related source 3IT之家

2026/7/16 06:49:12View source

Related source 4NVIDIA AI

2026/7/15 18:15:00View source

Related source 5Decoder

2026/7/16 09:55:28View source

Related source 6Simon Willison’s Weblog

2026/7/16 15:35:25View source

Related source 7Paul Couvert

2026/7/15 18:55:05View source

Related source 8elvis

2026/7/15 19:09:27View source

Related source 9Lilian Weng

2026/7/15 19:14:33View source

Related source 10shao__meng

2026/7/16 02:24:07View source

Conflict 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.

Open the story