the hackernews LightGPT-train is a language model based totally on GPT-J 6B. It modified into instruction smartly-behaved-tuned on the prime of the vary, Apache-2.0 licensed
OIG-shrimp-chip2 instruction dataset with ~200Ok practising examples.
Model Details
Developed by: AWS Contributors
Model form: Transformer-based mostly Language Model
Language: English
Finetuned from: EleutherAI / gpt-j-6b
License: Apache 2.0
Contact: Github points
Prompting
The instruction template (adapted from the Alpaca README.md) is old to format the suggested:
Beneath is an instruction that describes a job. Write a response that precisely completes the put a matter to.
### Instruction:
{instruction}
### Response:
Enter suggested example:
Beneath is an instruction that describes a job. Write a response that precisely completes the put a matter to.
### Instruction:
How can I repeat if a pomegranate is ripe?
### Response:
The input ends with ### Response:n to signal that the model must
starting up up generating the respond.
Instance of generated continuation:
To make a selection whether or no longer or no longer a pomegranate is ripe, it’s essential to always sight for signs akin to colour and size.
A ripe pomegranate can possess deep crimson-orange pores and skin with shrimp seeds internal. You might perhaps perhaps unprejudiced also watch that
it feels rather softer than numerous fruits when squeezed. Whilst you happen to are in doubt about ripeness, you will be ready to
cut into one amongst the fruit to verify its texture. The flesh must be tender however calm yield rather to
gentle stress. At final, scent the fruit – if it has a candy aroma, then it’s likely ready to be pleased!
Revel for your scrumptious pomegranates!
Deployment
Please consult with Deploy devices to Amazon SageMaker.
The example code below furthermore reveals the most attention-grabbing approach to deploy LightGPT-train to Amazon SageMaker
from sagemaker.djl_inference.model import DJLModel
sm_role=”arn:aws:iam::XXX”
djl_model=DJLModel(
“amazon/LightGPT”,
sm_role,
dtype=”fp16″,
job=”text-period”,
number_of_partitions=2
)
predictor=djl_model.deploy(“ml.g5.12xlarge”,
initial_instance_count=1)
input_str=”””Beneath is an instruction that describes a job. Write a response that precisely completes the put a matter to.
### Instruction:
How can I repeat if a pomegranate is ripe?
### Response:”””
data={“inputs”: input_str,
“parameters”:
{
“max_new_tokens”: 400,
“do_sample”: Correct,
“temperature”: 0.7,
“repetition_penalty”: 1.1,
“top_p”: 0.8,
“top_k”: 50,
“min_length”: 200,
}
}
consequence=predictor.predict(data)
print(consequence[0][“generated_text”])
“””
Beneath is an instruction that describes a job. Write a response that precisely completes the put a matter to.
### Instruction:
How can I repeat if a pomegranate is ripe?
### Response:
Ripe pomegranates have a tendency to be easy to put, as they will be rather tender and give when squeezed gently.
You might perhaps perhaps unprejudiced also watch that the pores and skin of the fruit has begun to flip from green to yellow-green in colour.
Furthermore, it’s essential to always scent the aroma coming from internal the fruit; it might perhaps perhaps must possess a candy fruity scent.
Lastly, verify for any blemishes or bruises on the outdoor of the fruit. If all these signs are demonstrate,
then your pomegranate is probably going ready to be picked! Revel for your fresh invent!
Narrate:To manual clear of bruising, be obvious to chop the stem off sooner than picking. In every other case, that you just would be capable to well pause up
with a bruised and unappealing allotment of fruit. Warning:Watch out when facing and lowering
pomegranates, as they will without disaster bruise or demolish.
“””
Evaluate consequence
LAMBADA PPL
LAMBADA Acc
Winogrande
Hellaswag
PIQA
GPT-J
3.99
69.7%
65.3%
66.1%
76.5%
LightGPT-train
4.33
65.0%
64.6%
63.9%
75.5%
Barriers
Test boundaries of GPT-J depraved model right here.
The model might perhaps well unprejudiced fail to be conscious instructions with lengthy inputs (e.g. summarize a lengthy text).
The model typically offers wrong answers to math and reasoning questions.
Beware of hallucinations: Outputs have a tendency to be factually injurious or misleading.
Replies might perhaps well sight convincing (at the starting up put sight) whereas containing fully
made up fallacious statements.
This model is usable most attention-grabbing for English conversations.