快讯
实测：RTX 3060 单卡跑 Liquid AI 开源模型 d1-3B，BoolQ 88.4%、MMLU-Pro 44.5%
作者称在单张 RTX 3060（12GB 显存）上运行 Liquid AI 新发布的开源多模态决策模型 d1-3B：BoolQ 3,270 题全量验证集得分 88.4%（高于官方卡片 86.7），MMLU-Pro 12,032 题得分 44.5%，GPQA 29.8% 接近随机，与其知识是最弱项的定位一致；约 43ms p50 每次决策、每秒 21-23 次决策、占 5GB 显存。推文并引用 Liquid AI：Open d1 发布 d1-3B（文本+视觉）与 d1-omni-600M（文本+图像或文本+音频）两个开源权重模型。
事件来源
查看原文
RT Doğukan running d1-3B on a single RTX 3060. 12GB VRAM. (config below) 88.4% on BoolQ (full val, 3,270 questions), Liquid's card says 86.7. 44.5% on MMLU-Pro (all 12,032, random is 10%). ~43 ms p50 per decision, 21-23 decisions/s, 5 GB VRAM. GPQA lands at 29.8%, close to random, which fits Liquid's own Decision Index showing knowledge as its weakest area. added a decision harness to BenchKit for this. calibration next! config (llama.cpp PR #30110, merged): llama-server -m d1-3B-Q8_0.gguf --mmproj mmproj-d1-3B-F16.gguf -ngl 99 -c 32768 -fa on -np 1 -b 4096 -ub 4096 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.