llama.cpp

Raised as a usability problem by 16 people on Hacker News over 4 years (2023–2026), in 13 threads.

People raised many different problems with llama.cpp, but no single problem has yet been raised by 3 people in 3 different months and threads.

Unreviewed issues were grouped by machine: Claude Haiku labelled each comment as a usability complaint, named what it's about and summed up the problem in a line, and a small model grouped similar problems. Nobody has checked them by hand. Expect mistakes.

The newest 16 complaints

Two settings.json files and unclear placement made setup too hard to figure out2026-08-23
Hard to find correct settings; quant and model variants are confusing for beginners2026-08-17
Hard to get running performantly; no clear instructions, slow results2026-08-11
Installing models seems nontrivial compared with Ollama's simple setup2026-07-20
Finding out how to download and install models was a pain2026-04-16
Built-in help is hard to navigate and parse2025-09-02
Needed manual compilation and hand-tuned layer settings to use the GPU well2025-01-31
Hard to run on Linux because of binfmt workarounds required2025-01-31
Hard to figure out how to use without digging; needs manual model downloads and configuration2025-01-31
Must look up recommended parameters and templates per model to get good results2025-01-12
Mingw build method endlessly fails with paths that never work right2024-11-30
Setup and daily use felt atrocious compared to alternatives2024-11-29
Hours spent getting the llama.cpp server running; choosing model, format and arguments is hard2024-06-16
Chat repeats input and re-prompts endlessly; image upload shows nothing happening2023-12-13
CPU output runs about one word per second, too slow for realtime chat2023-07-26
No simple-to-use interface to call the model from another app2023-04-18