How to speed up Deepseek v4.1 Flash max with Deepseek harness?

I am using Deepseek v4.1 flash max with deepseek harness. Most tasks are taking more than 35 minutes. For example I gave it task to migrate data from 2 old db tables which are modified in new database and create a standalone python script for migration by applying some transformations on data. I gave it DDL for old tables and it has access to current table structures in the codebase. I am finding same with most tasks, is there anything I may be doing wrong? Most other tasks are also taking 40 minutes or more. Cache hit for this task is at 99.6%, 29M Cached input and 165k output tokens. Edit: I asked it to analyze why this task took 35 minutes, some of time was wasted on windows powershell operations. For another 11 mins operation, it said 70% necessary and 30% because it expanded search space. Two self-inflicted hangs — Select-Object -First on unbuffered output, then ******* inheriting captured pipes on Windows 10–13 min Pure waste, 100% mineTwo self-inflicted hangs — Select-Object -First on unbuffered output, then ****** inheriting captured pipes on Windows 10–13 min Pure waste, 100% mine Analysis of another task which was also stuck: 1 Daemon-in-pipe hang (×3) ~20 min My bug, repeatable & avoidable 1 Daemon-in-pipe hang (×3) ~20 min My bug, repeatable & avoidable

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论