metadata

language:
  - en
datasets:
  - natural_instructions
  - the_pile
  - cot
  - Muennighoff/P3
tags:
  - ctranslate2
  - int8
  - float16 - gpt
pipeline_tag: text-generation
inference:
  parameters:
    temperature: 0.1
widget:
  - text: >-
      Is this review positive or negative? Review: Best cast iron skillet you
      will ever buy. Answer:
    example_title: Sentiment analysis
  - text: 'Where is Zurich? Ans:'
    example_title: Question Answering

# Fast-Inference with Ctranslate2

Speedup inference by 2x-8x using int8 inference in C++

quantized version of togethercomputer/GPT-JT-6B-v0

pip install hf-hub-ctranslate2>=2.0.6 ctranslate2>=3.13.0

Converted on 2023-05-19 using

ct2-transformers-converter --model togethercomputer/GPT-JT-6B-v0 --output_dir /home/michael/tmp-ct2fast-GPT-JT-6B-v0 --force --copy_files merges.txt tokenizer.json README.md tokenizer_config.json vocab.json special_tokens_map.json added_tokens.json .gitattributes --quantization float16

Checkpoint compatible to ctranslate2 and hf-hub-ctranslate2

compute_type=int8_float16 for device="cuda"
compute_type=int8 for device="cpu"

from hf_hub_ctranslate2 import TranslatorCT2fromHfHub, GeneratorCT2fromHfHub
from transformers import AutoTokenizer

model_name = "michaelfeil/ct2fast-GPT-JT-6B-v0"
# use either TranslatorCT2fromHfHub or GeneratorCT2fromHfHub here, depending on model.
model = GeneratorCT2fromHfHub(
        # load in int8 on CUDA
        model_name_or_path=model_name, 
        device="cuda",
        compute_type="int8_float16",
        tokenizer=AutoTokenizer.from_pretrained("togethercomputer/GPT-JT-6B-v0")
)
outputs = model.generate(
    text=["How do you call a fast Flan-ingo?", "User: How are you doing?"],
)
print(outputs)

Licence and other remarks:

This is just a quantized version. Licence conditions are intended to be idential to original huggingface repo.

Original description

Quick Start

from transformers import pipeline

pipe = pipeline(model='togethercomputer/GPT-JT-6B-v0')

pipe("Where is Zurich? Ans:")