Tooling / tuning
Training and tuning
Fine-tuning is a solved problem in the sense that the algorithms are published and the memory arithmetic is understood. These are the libraries that spare you implementing it — mostly wrappers around the same handful of methods, differing in how much they assume and how much they optimise.
4 entries.
Entries
09.02.1—09.02.2—09.02.3—09.02.4—
Config-Driven Fine-Tuningcommon
Axolotl, LLaMA-Factory and the case for a YAML file instead of a script.
PEFT and TRLstandard
The adapter library and the preference-training library, and how they compose.
Transformersstandard
The reference implementation of nearly every architecture, and the format everything reads.
Unslothcommon
Hand-written kernels and manual backward passes for fine-tuning on one GPU.