Repository navigation
Thunder Operations
pinokio edited this page Sep 28, 2026
·
2 revisions
Rules for running jobs on the remote Thunder GPU machines.
- Each machine runs one GPU job at a time.
- Check the process list and the job's log first.
Useful checks:
date '+%I:%M %p %Z'
ps -eo pid,ppid,etime,cmd
tail -f /home/ubuntu/<job>/eval.log- Never edit a shell script while it is running. Bash reads scripts as it goes, so an edit can make the running job jump to the wrong place. Write a new version under a new name and launch that instead.
- Stop jobs by PID, and only when a run is known to be invalid.
-
Never use
pkill -fwith a job pattern — it can match, and kill, the shell running it.
For a job that is still running, report the measured time per window, and keep it separate from the estimate of how much work is left.
Alexandria Audiobook2 · Repository · Issues
The user-guide pages started from the original Alexandria wiki and are kept up to date for this fork; the research pages document this fork's measured results.
Start here
Make an audiobook
Build voices
Reference
Research
- Which model for your card
- Results
- Prompts and adapters
- Evaluation recipes
- Thunder operations
- Hugging Face releases
日本語ガイド