|
--- |
|
license: other |
|
--- |
|
|
|
# WizardLM: An Instruction-following LLM Using Evol-Instruct |
|
|
|
These files are the result of merging the [delta weights](https://huggingface.co/victor123/WizardLM) with the original Llama7B model. |
|
|
|
The code for merging is provided in the [WizardLM official Github repo](https://github.com/nlpxucan/WizardLM). |
|
|
|
The original WizardLM deltas are in float32, and this results in producing an HF repo that is also float32, and so much larger than a normal 7B Llama model. |
|
|
|
Therefore for this repo I converted the model to float16, to produce a standard size 7B model. |
|
|
|
This was done achieved by running **`model = model.half()`** prior to saving. |
|
|
|
## WizardLM-7B HF |
|
|
|
This repo contains the full unquantised model files in HF format for GPU inference and as a base for quantisation/conversion. |
|
|
|
## Other repositories available |
|
|
|
* [4bit GGML models for CPU inference](https://huggingface.co/TheBloke/wizardLM-7B-GGML) |
|
* [4bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/wizardLM-7B-GPTQ) |
|
|
|
# Original model info |
|
|
|
## Full details in the model's Github page |
|
|
|
[WizardLM official Github repo](https://github.com/nlpxucan/WizardLM). |
|
|
|
## Overview of Evol-Instruct |
|
|
|
Evol-Instruct is a novel method using LLMs instead of humans to automatically mass-produce open-domain instructions of various difficulty levels and skills range, to improve the performance of LLMs. |
|
|
|
Although on our complexity-balanced test set, WizardLM-7B outperforms ChatGPT in the high-complexity instructions, it still lag behind ChatGPT on the entire test set, and we also consider WizardLM to still be in a baby state. This repository will continue to improve WizardLM, train on larger scales, add more training data, and innovate more advanced large-model training methods. |
|
|
|
 |
|
 |
