bytez 0.2.5

Python API client for Bytez service

API Documentation

Introduction

Welcome to the Bytez API documentation! This API provides access to various machine learning models for serverless operation. Below, you will find examples demonstrating how to interact with the API using our Python client library.

Getting Your Key

To use this API, you need an API key. Obtain your key by joining the Bytez Discord. If you prefer not to use Discord, email us at [e-mail].

Boot Times and Billing

Cold Boot Times

Expect the following boot times for models:

• Smallest model: ~12 minutes.
• Largest model: ~15 minutes. We are working on reducing these boot times to under 5 minutes.

Billing

Billing begins from the first 60 seconds of use, with subsequent usage rounded to the nearest minute. Charges are based on $0.0000166667 per GB-second on GPUs. The default expiration period for a model instance is 30 minutes.

Python Client Library Usage Examples

Authentication

Always include your API key when initializing the client:

from bytez import Bytez

client = Bytez('YOUR_API_KEY')

List Available Models

Lists the currently available models, and provides basic information about each one, such as RAM required

models = client.list_models()

print(models)

List Serverless Instances

List your serverless instances

instances = client.list_instances()

print(instances)

Make a Model Serverless

Make a HuggingFace model serverless + available on this API! Running this command queues a job. You'll receive an email when the model is ready.

@param modelId The HuggingFace modelId, for example openai-community/gpt2

model_id = 'openai-community/gpt2'

job_status = client.process(model_id)

print(job_status)

Get a Model

Get a model, so you can check its status, load, run, or shut it down.

@param modelId The HuggingFace modelId, for example openai-community/gpt2

model = client.model('openai-community/gpt2')

Start the model

Convenience method to model.start(), and then await for model to be ready.

@param options Serverless configuration

results = model.start('openai-community/gpt2', {'concurrency': 1, 'timeout': 300})

print(results)

# Concurrency
# Number of serverless instances.
#
# For example, if you set to `3`, then you can do 3 parallel inferences.
#
# If you set to `1`, then you can do 1 inference at a time.
#
# Default: `1`

# Timeout
# Seconds to wait before serverless instance auto-shuts down.
#
# By default, if an instance doesn't receive a request after `300` seconds, then it shuts down.
#
# Receiving a request resets this timer.
#
# Default: `300`

Check Model Status

Check on the status of the model, to see if its deploying, running, or stopped

status = model.status()

print(status)

Run a Model

Run inference

output = model.run("Once upon a time there was a")

print(output)

Run a Model with HuggingFace params

Run inference with HuggingFace parameters.

output = model.run("Once upon a time there was a", model_params={"max_new_tokens":1,"min_new_tokens":1})

print(output)

Stream the response

Streaming text

output = model.run("Once upon a time there was a", stream=True)

for chunk in stream:
  print(chunk)

Shutdown a Model

Serverless models auto-shutdown, though you can early stop with this method

model.stop()

Feedback

We value your feedback to improve our documentation and services. If you have any suggestions, please join our Discord or contact us via email.