就在刚刚,百度文心终于出手了,把自家基座模型文心 5.5 Flash Base 开源。
This is a dev post classified by AGK Intelligence as Other (a launch), kept by the Dev Radar because it carries real work, not commentary.
就在刚刚,百度文心终于出手了,把自家基座模型文心 5.5 Flash Base 开源。 一出手就在同尺寸模型里拿下 SOTA,在 30 项基准评测中表现突出。 超过了 DeepSeek-V4-Flash-Base、MiMo-V2.5-Base 等 Base 模型。 评测,覆盖了知识、推理、数学、代码、多语言和长上下文等方向。 模型开源地址:http://huggingface.co/collections/baidu/ernie-55-flash 模型从零开始预训练,喂了 33.5 万亿 Token,采用稀疏混合专家(MoE)架构。 总参数量 2800 亿,每个 Token 只激活其中的 140 亿,这样每次计算用到的参数少了,算力也省下不少。 另外上下文方面,预训练最长支持 128K,微调与推理支持 1M Token。 看了下它公开的技术拆解,花了不少心思在省算力上,比如 43 层里有 37 层把每 128 个 token 压成一条缓存,长文本就存得少了,感兴趣的朋友可以看下。 最后预判一下,基于它训练的文心 5.5 Flash 模型,估计很快就要发布了。 技术 Blog:http://ernie.baidu.com/zh/release/ernie-5-5-flash-base
Posted by GitHubDaily (91.4k followers) 57 min ago · 1 likes · 845 views · view the original post on X. Kept by the Dev Radar as Other. Tools mentioned: baidu, ERNIE 5.5 Flash Base.
More dev work like this
- 🚀Baidu just open-sourced ERNIE-5.5-Flash-Base — @Chinazhidx
- @Baidu_Inc released ERNIE 5.5 Flash Base. — @javierbits
- JUST IN: @VerneRobotics released V0.5, an open-weight model for industrial robots. — @IlirAliu_
- 读英文 PDF,最耗神的往往不是内容本身,而是术语和长句总把节奏打断。 — @bkdgiffug
- Nanjing Iron and Steel (NISCO) has 300+ AI models, 25 business agents, and 1,000+… — @Huawei
- 搭 3D 场景时缺几个小道具,桌上的杯子、墙角的椅子,自己建模太费时间,去素材站翻半天也不一定有合适的。 — @GitHub_Daily
- 2 minutes of continuous AI video with spoken dialogue, generated at home on two NVIDIA… — @majewskizby
- 油猴撑了好几年其实也够用,可真遇上页面关了脚本还得接着跑、定时任务又不想死盯标签页的活儿,它就有点接不住了。 — @bkdgiffug
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.7k posts from 5k X accounts over the last 21 days, 9.4k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-05 10:53 UTC. Full method.