Edit model card

Cheapity3 🐷

GPT-like T5 model trained to generate text in multiple languages.

Motivation

  • GPT models are expensive to run.
  • GPT models are monolingual.

Solution

  • Maybe, Small Models aren't Terrible (SMarT)
  • Plus, they are cheaper to run.

I fine-tuned T5 on multiple languages (πŸ‡¬πŸ‡§ English, πŸ‡©πŸ‡ͺ German, πŸ‡«πŸ‡· French) and multiple academic text snippets from various domains like tech, law, finance and science etc. to generate text, just like GPT models do.

Usage - NLPlayStore πŸ‘ˆ

from store.service_management import ServiceManager

service_manager = ServiceManager().get_service("cheapity3")

service.install()

service = service.launch()

input_text = "The mechanical engineering field requires ... "

generated_texts = service.play(input_text, 15) # A list a generated text

Usage - Hugging Face Transformers πŸ€—

  • Provide some text e.g "Italy, officially the Italian Republic is a country consisting of"
  • Tell Cheapity3 how many words you want to generate e.g 15 -- πŸ˜ƒ Yes, you can control the length.
  • Cheapity3 reads your text and generates a continuation containing approximately 15 words.
from transformers import AutoTokenizer, AutoModelWithLMHead

tokenizer = AutoTokenizer.from_pretrained("flexudy/cheapity3")

model = AutoModelWithLMHead.from_pretrained("flexudy/cheapity3")

input_text = """The mechanical engineering field requires an understanding of core areas including mechanics, dynamics,
 thermodynamics, materials science, structural analysis, and 
 electricity. { _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ }"""  # 15 words

inputs = tokenizer(input_text, return_tensors="pt", truncation=True, max_length=512)

input_ids = inputs["input_ids"]

attention_mask = inputs["attention_mask"]

outputs = model.generate(
    input_ids=input_ids,
    attention_mask=attention_mask,
    max_length=128,
    do_sample=True,
    early_stopping=True,
    num_return_sequences=4,
    repetition_penalty=2.5
)

for i in range(4):
    print(tokenizer.decode(outputs[i], skip_special_tokens=True, clean_up_tokenization_spaces=True))

INPUT: The mechanical engineering field requires an understanding of core areas including mechanics, dynamics, thermodynamics, materials science, structural analysis, and electricity.

> Cheapity3 continues with beam search:
... The field of mechanical engineering is a broad field that includes many core areas of engineering.
 
> Cheapity3 continues with sampling and top_k=50:
... Developing the knowledge base for these core areas will enable engineers to build their capabilities rapidly and efficiently. ...
... The field of mechanics offers a variety and broad range for applications throughout the engineering/technological fields. ...
... Mechanics generally is not understood by students. While they can be employed in the field, mechanical engineering ...
... Introduction to mechanical engineering and core fields including chemical products, materials science, structural analysis, and geomatics ...

Pretty decent right?

Hence, whenever you feel like GPT3 is too expensive, Cheapity3 comes to the rescue πŸ€—.

Model Training FYI

  • T5-base model
  • Trained on ONLY 1M sentences from English, French and German text
  • Mostly text from Wikipedia, arxiv and QA datasets
  • Learning rate: 0.00003
  • 2 epochs
  • Max input: 512 tokens
  • Max output: 128 tokens
Downloads last month
11
Hosted inference API
This model can be loaded on the Inference API on-demand.