加载中...

MP

MPT-30B-Instruct

Name: MosaicML Pretrained Transformer - 30B Instruct
Author: MosaicML

MosaicML Pretrained Transformer - 30B Instruct

Release date: 2023-06-22更新于: 2023-06-23 20:48:53.524117

Live demo

GitHub

Hugging Face

Compare

Parameters

300.0亿

Context length

2K

Chinese support

Not supported

Reasoning ability

Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

MPT-30B-Instruct

Model basics

Reasoning traces

Not supported

Context length

2K tokens

Max output length

No data

Model type

基础大模型

Release date

2023-06-22

Model file size

60

MoE architecture

No

Total params / Active params

300.0B / N/A

Knowledge cutoff

No data

Inference modes

No mode data

MPT-30B-Instruct

Open source & experience

Code license

Apache 2.0

Weights license

CC-BY-NC-SA-3.0- 不可以商用

GitHub repo

https://github.com/mosaicml/llm-foundry

Hugging Face

https://huggingface.co/mosaicml/mpt-30b-instruct

Live demo

No live demo

MPT-30B-Instruct

Official resources

Paper

MPT-30B: Raising the bar for open-source foundation models

DataLearnerAI blog

No blog post yet

MPT-30B-Instruct

API details

API speed

No data

No public API pricing yet.

MPT-30B-Instruct

Benchmark Results

No benchmark data to show.

MPT-30B-Instruct

Publisher

MosaicML

View publisher details

MosaicML Pretrained Transformer - 30B Instruct

Model Overview

MPT-30B-Instruct是MosaicML开源的MPT-30B系列中针对指令进行微调的大模型。该模型代码依然采用Apache 2.0开源协议开源，但是其预训练结果却是CC-BY-NC-SA-3.0协议，不允许商用。

MPT-30B是MosaicML训练的300亿参数规模的大模型，其基础模型相比较7B版本有了很大的改进，在效果和训练推理效率上都有很大的提升。

MPT-30B-Instruct通过在Dolly HHRLHF上对MPT-30B进行微调而建立的，Dolly HHRLHF来源于Databricks Dolly-15k和Anthropic Helpful and Harmless（HH-RLHF）数据集。它还在Competition Math、Duorc、CoT GSM8k、Qasper、Quality、Summ Screen FD和Spider上训练。

Foundation model

MPT

View details

DataLearner 官方微信

欢迎关注 DataLearner 官方微信，获得最新 AI 技术推送