Create and parse prompts for large language models.
Project description
AIPrompts
This is a simple prompt builder for OpenAI models. Easy to use and to modify.
Install
pip install AIPrompts
pip install AIPrompts@git+https://github.com/TeiaLabs/prompts.git@master
Dynamic prompts
template = 'a photo of a <img_label>'
prompt = DynamicPrompt(template)
filled_prompt = prompt.build(img_label='dog')
print(filled_prompt)
# out: "a photo of a dog"
Dynamic prompts from file templates
Build your own prompt by creating a file following a sample.prompt file (yaml format), and use the DynamicPrompt class to parse it:
prompt = DynamicPrompt.from_file('samples/sample.prompt')
str_prompt = prompt.build(
input_sentence="lets go",
)
You can also access recommended model settings (engine, temperature) that can be fed to the model input (e.g., openai.Completion.create()):
prompt.get_model_settings()
Improve Autocomplete with custom prompts
Alternatively, to get more control and better autocomplete suggestions, you can inherit from the BasePrompt
class and override the build method with explicit arguments:
class MyPrompt(BasePrompt):
def build(self, input_sentence):
return self.set_prompt_values(
input_sentence=input_sentence,
)
Ensembling prompts
To ensemble multiple prompts, you can use the EnsemblePrompt
class:
templates = [
'<label>',
'a photo of <label>',
'picture of <label>',
]
exp_vars = ['label']
prompt = PromptEnsemble(templates, exp_vars)
prompt.build(label='dog')
# out: ['dog', 'a photo of dog', 'picture of dog']
prompt.build(label='cat')
# out: ['cat', 'a photo of cat', 'picture of cat']
The output is a flattened list with all filled in templates. Note: all templates must be filled with the same expected variables, and all variables must be provided.
You can also build multiple promtps at the same time (useful for classification):
templates = [
'<label>',
'a photo of <label>'
]
template_vars = [
'label'
]
labels = ['dog', 'cat', 't-shirt']
prompt = PromptEnsemble(templates, template_vars)
prompt.build_many(
label=labels
)
# out: ['dog', 'a photo of dog', 'cat', 'a photo of cat', 't-shirt', 'a photo of t-shirt']
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for AIPrompts-0.2.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | d5bb805eb7d407b24988449a44d53bb305dec592cd9871e4de60fd50ae248db4 |
|
MD5 | aeeb411ab4f2a109633164b3df8e70e6 |
|
BLAKE2b-256 | 3a4421ade89011cc86b1a368bc68e3e444c8163737881b34abcfcc6411e266d1 |