GPTQ
Also known as: GPTQ quant
GPTQ is a post-training quantization method that compresses a model to 4 bits with limited quality loss.
- Mature tooling (AutoGPTQ / GPTQModel).
- Works across many models.
- A common alternative to AWQ.
The landscape: GGUF/K-quants dominate local CPU+GPU use; AWQ and GPTQ dominate fast GPU serving. All three are "4-bit-ish" — the differences are subtle.