跳到正文
菜单
简体中文

推荐技术栈

你在构建什么?Customise a model

每个位置一项技术,并说明推荐理由以及可以改用什么。

Base model

model

GPT

Family of decoder-only Transformer language models trained with a generative objective.

Why 由图谱推导

The agents converse through a model backend, with the hosted models among the supported ones.

来源

Method

technique

Fine-tuning

Continue training a pre-trained model on task-specific data to adapt its behaviour.

Why 由图谱推导

LoRA used for Fine-tuning

来源

Efficient alternative

technique

LoRA

Fine-tuning by training small low-rank adapter matrices instead of all weights.

Why 由图谱推导

QLoRA adds 4-bit quantisation of the frozen base model on top of LoRA adapters.

来源

Serving

technique

KV Cache

Cached attention keys and values that make generating one token at a time affordable.

Why 由图谱推导

Reusing a shared prefix means reusing the cached keys and values for it.

来源