Few-Shot Learning VS Simple Instructions
When the first large language models were released, like GPT-J, OPT, Bloom, etc. it quickly appeared that - despite being very powerful - these models were not able to understand simple human instructions made in natural language.
For example, if you want to extract a name, a position, and a company, from a piece of text, you need to do something like this using GPT-J on NLP Cloud:
import nlpcloud
client = nlpcloud.Client("gpt-j", "your_token", gpu=True)
generation = client.generation("""[Text]: Fred is a serial entrepreneur. Co-founder and CEO of Platform.sh, he previously co-founded Commerce Guys, a leading Drupal ecommerce provider. His mission is to guarantee that as we continue on an ambitious journey to profoundly transform how cloud computing is used and perceived, we keep our feet well on the ground continuing the rapid growth we have enjoyed up until now.
[Name]: Fred
[Position]: Co-founder and CEO
[Company]: Platform.sh
###
[Text]: Microsoft (the word being a portmanteau of "microcomputer software") was founded by Bill Gates on April 4, 1975, to develop and sell BASIC interpreters for the Altair 8800. Steve Ballmer replaced Gates as CEO in 2000, and later envisioned a "devices and services" strategy.
[Name]: Steve Ballmer
[Position]: CEO
[Company]: Microsoft
###
[Text]: Franck Riboud was born on 7 November 1955 in Lyon. He is the son of Antoine Riboud, the previous CEO, who transformed the former European glassmaker BSN Group into a leading player in the food industry. He is the CEO at Danone.
[Name]: Franck Riboud
[Position]: CEO
[Company]: Danone
###
[Text]: David Melvin is working for CITIC CLSA with over 30 years’ experience in investment banking and private equity. He is currently a Senior Adviser of CITIC CLSA.
""",
top_p=0,
length_no_input=True,
end_sequence="###",
remove_end_sequence=True,
remove_input=True)
print(generation["generated_text"])
This technique, known as "few-shot learning", or "prompt engineering" is explained in a dedicated article: read the article here.
Few-shot learning works very well on ChatGPT and ChatDolphin and allows you to get very advanced results. But in most cases few-shot learning is not needed and unnecessarily complex. Besides, as the generative AI models only allow for a limited input length, the few-shot examples sometimes simply don't fit into the request.
Good news is that, when properly fine-tuned, the large language models can learn how to understand human instructions without using few-shot learning. This is the case of ChatGPT and ChatDolphin.
With these models, here is how your query would look like:
import nlpcloud
client = nlpcloud.Client("chatdolphin", "your_token", gpu=True)
generation = client.generation("""Extract name, position, and company, from the following text.
David Melvin working for CITIC CLSA with over 30 years’ experience in investment banking and private equity. He is currently a Senior Adviser of CITIC CLSA.""")
print(generation["generated_text"])
Output:
Name: David Melvin
Position: Senior Adviser
Company: CITIC CLSAMuch simpler isn't it? Now what if we want the result to be formatted as JSON? Here is a simple instruction:
import nlpcloud
client = nlpcloud.Client("chatdolphin", "your_token", gpu=True)
generation = client.generation("""Extract name, position, and company, from the following text. Format the result as JSON.
David Melvin working for CITIC CLSA with over 30 years’ experience in investment banking and private equity. He is currently a Senior Adviser of CITIC CLSA.""")
print(generation["generated_text"])
Output:
{
"name": "David Melvin",
"position": "Senior Adviser",
"company": "CITIC CLSA"
}I think you get the idea don't you?
You can easily test natural instructions on the NLP Cloud Playground, in the text generation section. Click here to try text generation on the Playground. Then simply use one of the examples showed below in this article and see for yourself.
Note that these models are trained to generate large responses. If you need short and concise responses, you can mention it in your prompt (with something like "Make a short response.").
Entity extraction example with ChatDolphin on the NLP Cloud Playground