Model Card for FLAN T5 XL Q8

The model is quantized version of the google/flan-t5-xl with int8 quantization.

Model Details

Model Description

The model being quantized using CTranslate2 with the following command:

ct2-transformers-converter --model google/flan-t5-xl --output_dir google/flan-t5-xl-ct2 --copy_files tokenizer.json tokenizer_config.json special_tokens_map.json spiece.model --quantization int8 --force --low_cpu_mem_usage

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

pip install -qU ctranslate2 transformers[torch] sentencepiece accelerate

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

translator = ctranslate2.Translator("google/flan-t5-xl-ct2")
tokenizer = transformers.AutoTokenizer.from_pretrained("google/flan-t5-xl-ct2")

input_text = "translate English to German: The house is wonderful."
input_tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(input_text))

results = translator.translate_batch([input_tokens])

output_tokens = results[0].hypotheses[0]
output_text = tokenizer.decode(tokenizer.convert_tokens_to_ids(output_tokens))

print(output_text)

The code is taken from https://opennmt.net/CTranslate2/guides/transformers.html#t5.

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