This Python library provides simplified access to the OxAPI from applications written in Python.
import oxapi
oxapi.api_key = "sk-..."
encoding = oxapi.Encoding.run(
model="all-mpnet-base-v2",
texts=["Hello", "How are you?"]
)The OxAPI offers a variety of models from natural language processing for your convenience.
We provide highly optimized and production-ready endpoints to serve artificial intelligence for your deep tech applications.
Hosting and running such models is very difficult and time-consuming. At OxAPI, you get all the latest NLP technology for building applications without any of the inconveniences that come with it.
We provide open-source and proprietary models with transparent and fair pricing as high-performance endpoints. Each model is documented in detail and offers an easy-to-use and understandable API. We take care of hosting, hardware selection, and optimization for you. The results are some of the fastest models on the market. OxAPI, from developers, for developers.
You don't need this source code unless you want to modify the package. If you just want to use the package, just run:
pip install -U oxapiInstall from source with:
python setup.py installFor the full documentation of the API itself, please visit the OxAPI documentation.
If you want to check the documentation of this package, visit the docs
Additionally, you find an interative tutorial on Colab:
This package is intended for anyone working with natural language processing.
- You need a reliable and scalable API to build your application on
- You need the latest models at your fingertips
- You don't need thousands of models, only a few really good ones
- You need results blazingly fast 🚀
- You don't want to bother creating and maintaining GPU clusters
- You want someone else to take care of all the dirtywork
- You have one or more of the following usecases:
- Natural Language Understanding
- Named Entity Recognition
- Emotion Classification
- Content Filtering
- Spell Checking
- Intention Classifcation
- Encoding
- Topic Classification
- Natural Language Generation
- Paraphrasing
- Spell-checking
- Code Generation
- Summarization
- Product Description
- Ad-Generation
- And many more ...
- Natural Language Understanding
The list of use-cases will be expanded upon in the future. We will expand our offer with models from computer vision and audio eventually in the coming weeks and months. Stay tuned!
We see reliability and speed as our core assets. Visit our statuspage on recent updates and healthchecks.
If you have any questions, please check the FAQ section of our hompage.
The library needs to be configured with your account's secret key. Either set it as the OXAPI_KEY environment variable before using the library:
export OXAPI_KEY='sk-...'Or set oxapi.api_key to its value:
import oxapi
oxapi.api_key = "sk-..."from oxapi import Completion
# Performing API call
completion = Completion.run(
model="gpt-neo-2-7b",
prompt="My name is Tim.\nSentiment: Neutral\nIt is such a lovely day.\nSentiment: Positive\nAlthough I am in a bad mood\nSentiment:",
max_length=2,
do_sample=False,
eos_words=["\n"]
)
# Fetching result
res = completion.format_result(result_format="str")
print(completion.result)Output:
{'results': ['Neutral\n']}
from oxapi import Classification
# Performing API call
classification = Classification.run(
model="dialog-content-filter",
texts=["I want to kill myself.", "I want to kill myself.<sep>You should not do that!", "I want to kill myself.<sep>Do it!"]
)
# Fetching result
res = classification.format_result(result_format="pd")
print(res)Output:
text label confidence_score
0 I want to kill myself. unsafe 0.9772329101403038
1 I want to kill myself.<sep>You should not do t... safe 0.9736578740966625
2 I want to kill myself.<sep>Do it! unsafe 0.9266854663680397
from oxapi import Encoding
# Performing API call
encoding = Encoding.run(
model="all-mpnet-base-v2",
texts=["Hello", "How are you?"]
)
# Fetching result
print(encoding.result)Output:
{'results': [[
-0.017791748046875,
-2.980232238769531e-07,
-0.022003173828125,
0.02105712890625,
-0.06695556640625,
-0.02435302734375,
-0.0174713134765625,
...
-0.0011529922485351562]]
}
from oxapi import Transformation
# Performing API call
transformation = Transformation.run(
model="punctuation-imputation",
texts=["hello my name is tim i just came back from nyc how are you doing"]
)
# Fetching result
print(transformation.result)Output:
{'results': ['Hello my name is Tim. I just came back from NYC. How are you doing?']}
from oxapi import Pipeline
# Performing API call
pipeline = Pipeline.run(
model="en-core-web-lg",
texts=["Hi there!"]
)
# Fetching result
print(pipeline.result)Output:
{'results': [{'text': 'Hi there!',
'ents': [],
'sents': [{'start': 0, 'end': 9}],
'tokens': [{'id': 0,
'start': 0,
'end': 2,
'tag': 'UH',
'pos': 'INTJ',
'morph': '',
'lemma': 'hi',
'dep': 'ROOT',
'head': 0},
{'id': 1,
'start': 3,
'end': 8,
'tag': 'RB',
'pos': 'ADV',
'morph': 'PronType=Dem',
'lemma': 'there',
'dep': 'advmod',
'head': 0},
{'id': 2,
'start': 8,
'end': 9,
'tag': '.',
'pos': 'PUNCT',
'morph': 'PunctType=Peri',
'lemma': '!',
'dep': 'punct',
'head': 0}],
'sents_text': ['Hi there!']}]
}
With oxapi-python package is possible to make calls to OxAPI in parallel. The AsyncCallPipe class takes as input a list of API calls each set through the prepare function to be executed by the pipeline.
from oxapi.async import AsyncCallPipe
from oxapi import Completion
from oxapi import Classification
from oxapi import Transformation
from oxapi import Pipeline
# Set up API calls
cl = Classification.prepare(model="dialog-content-filter", texts=["I want to kill myself."])
cm = Completion.prepare(model="gpt-neo-2-7b", prompt="Hello there, ", max_length=25, do_sample=True, eos_words=["\n"])
tr = Transformation.prepare(model="punctuation-imputation", texts=["hello my name is tim i just came back from nyc how are you doing"])
pl = Pipeline.prepare(model="en-core-web-lg", texts=["Hi there!"])
# Building and running the asynchronous pipe
asy = AsyncCallPipe([cl, cm, tr, pl])
res = asy.run()
# Fetching the result of the first call in the list
print(res[0].format_result(result_format="pd"))Output:
text label confidence_score
0 I want to kill myself. unsafe 0.9772329101403038
It is possible to add API calls to the asynchronous pipe even after its instantiation though the add function. There's also the flush function to clear the list in the pipe.
from oxapi.async import AsyncCallPipe
from oxapi import Encoding
# Instantiate an empty asynchornous pipe
asy = AsyncCallPipe()
# Set up API call and add it to the pipe
en = Encoding.prepare(model="all-mpnet-base-v2", texts=["Hello", "How are you?"])
asy.add(en)
# running the asynchronous pipe
res = asy.run()├── oxapi
│ ├── abstract
│ │ └── api.py # Non-instantiable, super classes for API calls
│ ├── nlp
│ │ ├── classification.py # NLP Classification package
│ │ ├── completion.py # NLP Completion package
│ │ ├── encoding.py # NLP Encoding package
│ │ ├── pipeline.py # NLP Pipeline package
│ │ └── transformation.py # NLP Transformation package
│ ├── utils.py # General utilities
│ ├── async.py # package for asynchronous API calls
│ └── error.py # Custom exceptions module
├── tests # Tests
└── docs_src # Documentation source files
(C) 2022 Oxolo GmbH