返回 2026-08-17 编辑版
发现与解读中文优先 · EN 原文推理基础设施

Lamb Labs (YC S26) 发布:250 美元的 FPGA 板上可达到 12,000 tok/s。

250 美元的 FPGA 板上可达到 12,000 tok/s。这是一个小模型,并不能连贯地回答任何问题或提示。在 https://trylittlelamb.com/ 上尝试一下,但不要指望能聊得很好。至少它非常快

查看英文原文

12,000 tok/s on a 250$ FPGA board. It's a small model, and doesn't answer coherently to any question or prompt really. try it out on https://trylittlelamb.com/ but don't expect a good chat. at least its VERY fast

12,000 tok/s on a 250$ FPGA board. It's a small model, and doesn't answer coherently to any question or prompt really. try it out on https://trylittlelamb.com/ but don't expect a good chat. at least its VERY fast

本地模型不再只比能否运行,速度、上下文和回答质量必须放在同一张账上

为什么值得看

模型下沉到本地设备后,产品判断不能只看峰值速度或权重是否可下载。硬件成本、上下文长度、任务质量和交互体验必须一起核算,才知道一种部署形态是否真的适合工作流。

最强反方

FPGA 吞吐、模型规模与本地体验分别来自项目方、模型页和个人评测,测试任务与设备条件不一致;这些信号证明可运行形态在扩展,但不能组成统一性能排名。

原始出现与传播对照

1 条出现 · 本次 AIHOT 未精确命中

发现与解读

250 美元的 FPGA 板上可达到 12,000 tok/s。这是一个小模型,并不能连贯地回答任何问题或提示。在 https://trylittlelamb.com/ 上尝试一下,但不要指望能聊得很好。至少它非常快

250 美元的 FPGA 板上可达到 12,000 tok/s。这是一个小模型,并不能连贯地回答任何问题或提示。在 https://trylittlelamb.com/ 上尝试一下,但不要指望能聊得很好。至少它非常快

查看英文原文

12,000 tok/s on a 250$ FPGA board. It's a small model, and doesn't answer coherently to any question or prompt really. try it out on https://trylittlelamb.com/ but don't expect a good chat. at least its VERY fast

12,000 tok/s on a 250$ FPGA board. It's a small model, and doesn't answer coherently to any question or prompt really. try it out on https://trylittlelamb.com/ but don't expect a good chat. at least its VERY fast

直接发布标题Lamb Labs (YC S26)@LambLabs打开该层出处 ↗

12,000 tok/s on a 250$ FPGA board. It's a small model, and doesn't answer coherently to any question or prompt really. try it out on https://trylittlelamb.com/ but don't expect a good chat. at least its VERY fast

来源
X @ycombinator · x-ycombinator
证据角色
x_project_network
时间状态
known
用户判决
未判
查看该来源原文raw/x-ycombinator.jsond7fd34426b3d3984510b5bc462c0c02bf818e0959bca008bf91aabe602aa3d4c
Cluster IDevt_b07457299920989c881e1c4e精确去重依据canonical_url事件状态未判