RCreddit.com
18
·开发者社区 · RSS
Shots were fired
这条记录涉及生成能力或端侧推理进展，适合跟踪模型效率、部署门槛和应用机会。
为了节省成本，我使用本地大型语言模型（Qwen 3.6 35B）对项目进行基本维护，包括重新启动管道以使其在特定点恢复。当Qwen最初遇到困难但最终完成任务后，我要求它将操作方法添加到文档中，以避免将来再次遇到问题。后来，我与Claude合作调试管道失败的原因，以防止未来发生类似情况。在这次调试中，我提到一个“低端LLM”曾负责恢复管道。…
To save cost, I had a local LLM (qwen 3.6 35B) execute some basic maintenance on a project (relaunch a pipeline to resume it at a specific point). When Qwen struggled to do it, but eventually figured it out, I asked it to add to the documentation the method it used to do the operation so that it wouldn't struggle next time.
Later on, I worked with Claude to debug WHY the pipeline failed so that we could prevent it in the future; I mentioned that a "lower-end LLM" had to resume the pipeline.
主题标签ClaudeQwen模型发布端侧推理
原始关键词#fired#shots