MPT-7B-Instruct Q8

The model is quantized version of the mosaicml/mpt-7b-instruct with int8 quantization.

Model Details

Model Description

The model being quantized using CTranslate2 with the following command:

ct2-transformers-converter --model mosaicml/mpt-7b-instruct --output_dir mosaicml/mpt-7b-instruct-ct2 --copy_files tokenizer.json tokenizer_config.json special_tokens_map.json generation_config.json --quantization int8 --force --low_cpu_mem_usage --trust_remote_code

If you want to perform the quantization yourself, you need to install the following dependencies:

pip install -qU ctranslate2 transformers[torch] accelerate einops

Shared by: Lim Chee Kin
License: Apache 2.0

How to Get Started with the Model

Use the code below to get started with the model.

import ctranslate2
import transformers

generator = ctranslate2.Generator("limcheekin/mpt-7b-instruct-ct2")
tokenizer = transformers.AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b")

prompt = "Long long time ago, "
tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(prompt))

results = generator.generate_batch([tokens], max_length=256, sampling_topk=10)

text = tokenizer.decode(results[0].sequences_ids[0])

The code is taken from https://opennmt.net/CTranslate2/guides/transformers.html#mpt.

The key method of the code above is generate_batch, you can find out its supported parameters here.