RCreddit.com
12
·21小时前·开发者社区 · RSS
My RULE of Thumb of choosing a models
Qwen模型发布开源代码
热度趋势
新上榜
百分比基于当前可用热度信号,而非评论数或独立用户人数。
Qwen 相关模型动态已经出现,适合跟踪能力变化、生态影响和后续可用性。
一位开发者分享了选择模型的心得,指出大型语言模型(LLM)显著缩短了开发时间。例如,在没有LLM的情况下,调试或实现一个功能可能需要15小时,但使用Qwen 27B(甚至在Qwen 3.8之前)只需4小时。开发者还提到,0.5 tok/s的处理速度与人类速度相当,这使得进行通宵的代码库分析、金融科技或深度研究成为可能。
This is mostly for setting up for expectation, since personally without LLM i could take 3 days (15 hours of active programming) to debug or implement a feature, but with Qwen 27B (even before Qwen 3.8) it take 4 hours.
And yes 0.5 tok/s is human, not accounting of deletion and pausing, that's also the reason i am fine leaving overnight code base wide analysis or fin tech and deep research.