|
|
|
|
|
|
by antonvs
257 days ago
|
|
|
No-one on LocalLlama is training their own models. They’re working with foundation models like Llama from Meta and tweaking them in various ways: fine tuning, quantizing, RAG, etc. There’s a limit to how much improvement can be made like that. The basic capabilities of the foundation model still constrain what’s possible. |
|
|