Uses the 'Evol-Instruct' method to progressively rewrite simple instructions into increasingly complex and diverse ones using GPT-4. Results in a dataset where ~70k examples span from basic to graduate-level difficulty. WizardLM 13B outperformed Vicuna 13B on complex reasoning tasks using this data alone.
| Provider | WizardLM |
| Category | 指令 / SFT |
| Size | 70k Rows |
| License | Apache 2.0 |
| Downloads | 800k |
| Tags | Synthetic, Evol-Instruct, Complex-Reasoning, GPT-4 |
from datasets import load_dataset
ds = load_dataset("WizardLM/WizardLM_evol_instruct_70k")
使用 QLoRA(4-bit 基础模型 + LoRA 适配器)微调的预计显存需求(保守默认参数):
| 7B QLoRA | ~6GB VRAM |
| 13B QLoRA | ~10GB VRAM |
微调新手?跟着分步教程走: 一小时微调你的第一个 LLM