sentinelKV2EX member #631792, joined on 2023-05-30 20:41:02 +08:00Today's activity rank 14181 |
| 你好 NVFP4: strata-0.1.44.1, qwen3.8-flash-next, rtx pro 5000 48GB + 64GB RAM, decode 130t/s, prefill2600 t/s Local LLM • sentinelK • 9 days ago • Lastly replied by panghu960 | 8 |
| strata 极致优化指南(参数篇)64GB RAM + RTX pro 5000 48GB, IQ3_XXS 量化, prefill 4500t/s, decode 180t/s 同时缓存 5 个 262k 会话不重新 prefill Local LLM • sentinelK • 11 days ago • Lastly replied by Ken1028 | 15 |
| “马鞍税”,一篇非常有意思的 harness 横比文章 程序员 • sentinelK • Sep 18 • Lastly replied by greentim | 9 |
| 很好奇小米开放的这个实时训练监控器,其中成本(cost)的组成: https://mimo.xiaomi.com/rl/#overview 程序员 • sentinelK • Sep 17 • Lastly replied by lwep | 8 |
| deepseek-flash 的第三方跑分趣闻 DeepSeek • sentinelK • Sep 11 • Lastly replied by sentinelK | 3 |
| Hermes Agent 9 月更新纪实: bot 机制补全了 hermes 的最后一块基石,产品功能设计逐渐向 GUI 倾斜。 Hermes Agent • sentinelK • 12 days ago • Lastly replied by peacekang | 1 |
| mac m5 ultra 上线中国官网。1.2T 显存带宽, 256GB 版本 8.6 万 Local LLM • sentinelK • Aug 28 • Lastly replied by sadan9 | 51 |
| RTX PRO 5000(或其他 48GB 显存)的 Qwen3.8-27B-FP8 配置交流(prefill 5000+t/s, decode 60+t/s)
1 Local LLM • sentinelK • Aug 26 • Lastly replied by tgfbeta
|
55 |
| OpenCode go 订阅开始区分模型提供额度, kimi-k3、deepseek-v4-pro 等只有 15$的额度,请各位注意用量。 程序员 • sentinelK • Jul 20 • Lastly replied by shuck | 49 |
1 day ago Replied to a topic by coderMonkey › 程序员 › 好奇大家平时都是用的啥 agent,还有就是用桌面端还是 cli,这两者有区别吗 |
2 days ago Replied to a topic by wyl986 › 程序员 › 全程家宽+软路由+全局,还是被 Claude 发现来自大陆 |
3 days ago Replied to a topic by formulahendry › Visual Studio Code › The Story of VS Code | 官方纪录片 |
4 days ago Replied to a topic by 6581 › 程序员 › 虽然 AI 编码很厉害,但是这些问题怎么办呢? |
8 days ago Replied to a topic by sunxfancy › 程序员 › AI 在有审美的情况下强的可怕 |
8 days ago Replied to a topic by sunxfancy › 程序员 › AI 在有审美的情况下强的可怕 |
8 days ago Replied to a topic by doujiangjiyaozha › 职场话题 › 上午刚给自己项目开源,下午就收到 hr 裁员,卧槽了,我还没发育完呢 |
9 days ago Replied to a topic by tes286 › 程序员 › 如果在让 ai 做网页的时候指定设计系统,然后手写文案,能否比较有效的降低 ai 味? |
9 days ago Replied to a topic by Chihiro316 › 程序员 › 我用 deepseek 逆向了小米运动的卡路里算法,绷不住了 |
10 days ago Replied to a topic by liuluxu1989 › 程序员 › 刚 Tibo 说 codex 支持会话的永久缓存,有懂技术的解释下这可能吗 |