← 返回快讯

快讯

Liquid AI 发布 Open d1:两个开源多模态决策模型

liquidai

Liquid AI 宣布发布 Open d1,推出 d1 决策模型家族的两个开源权重多模态模型:d1-3B 支持文本加视觉,d1-omni-600M 支持文本加图像或文本加音频。官方称模型面向实时决策,可部署于从 NVIDIA DGX 数据中心、RTX 工作站到 Jetson 边缘设备等多种环境。转发者表示 d1-3B 在 RTX 4090 上据称延迟为 8 毫秒,并称其为"一次前向传播输出结果、零输出 token"的决策模型;这些性能数据来自作者转述,未经独立核实。

查看原文
RT Sudo su another decision model family, liquid's d1, and these are tiny > d1-3b, text and vision, liquid says 8 ms on a 4090 > d1-omni-600m, text plus image or audio, no latency published yet > one forward pass, odds over fixed answers, zero output tokens > lfm open license, free under $10m revenue clef flash 9b played breakout for me at 56 ms a move, 18.4 saves per ball, the 27b just sat under an idle paddle a 3b at 8 ms should hold that paddle better, same harness, same game, i'll run both Liquid AI: Today we release Open d1: two open-weight multimodal models in our d1 decision model family. > d1-3B: text + vision > d1-omni-600M: text + image or text + audio > Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.

前后快讯