Fine-tuning llama.cpp is costly and unpredictable
MLOps2 posts from 2 people2d active+11 post in the last 7 days, 1 the 7 days before (steady)Unverified
Posts per day
The posts behind it
2, newest first| Post | Date | |
|---|---|---|
| Feature Request: multi_logit_bias...to logit_bias that allows me to blacklist phrases ### Motivation it's getting more difficult to make good anti-slop finetunes, so i figured why not do it here instead? generate a bunch of text, look for overused bigrams/trigrams (what i usually do when making a dataset for this)...ggml-org/llama.cppelectroglyph | ggml-org/llama.cppelectroglyph4 reactions, 4 replies | Oct 32 days ago |
| ...their first release. BTL-2 (or -3 maybe? I forget) looked kind of interesting but it was annoying to run, needed a patch on llama.cpp, looks like they're improving their tooling. I'm certainly interested in how this actually performs.Hacker News commentskadoban | Hacker News commentskadoban | Sep 2213 days ago |
Companies and products named
About this problem
Evidence
2 posts from 2 people in 2 places, about 1 a week over 12 days. Mostly on Hacker News comments, ggml-org/llama.cpp. Tools named alongside: GitHub.
Frustration Frustration 0 of 3· Seen on Hacker News, GitHub issues
How it was grouped
Posts that state a pain and match the "fine-tuning" rule about llama.cpp. First post Sep 22, 2026, latest Oct 3, 2026. Unverified: fewer than 3 posts, or one person or place only. Method
The same problem elsewhere
Other llama.cpp problems
History
- 2026-10-05 Added: Fine-tuning llama.cpp is costly and unpredictable (2 posts)