Skip to content

Latest commit

 

History

History
224 lines (178 loc) · 5.27 KB

README.md

File metadata and controls

224 lines (178 loc) · 5.27 KB
Raw

Remyx AI command-line client

Installation

To install the Remyx AI CLI in Python virtual environment, run:

pip install remyxai

Token authentication

Remyx AI API requires authentication token, which can be obtained on this page: https://engine.remyx.ai/account

Provide api key to the CLI through an environment variable REMYXAI_API_KEY.

export REMYXAI_API_KEY=<your-key-here>

Usage

Quickly get started with the following examples:

Model

List all models:

  • cli command:
$ remyxai model list
  • python command:
from remyxai.api import list_models
print(list_models())

Get the summary of a model:

  • cli command:
$ remyxai model summarize --model_name=<your-model-name>
  • python command:
from remyxai.api import get_model_summary
print(get_model_summary(model_name))

Delete a model by name:

  • cli command:
$ remyxai model delete --model_name=<your-model-name>
  • python command:
from remyxai.api import delete_model

model_name = "<your-model-name>"
print(delete_model(model_name))

Download and convert a model:

  • cli command:
# possible model formats are "blob", "onnx", or "tflite"
$ remyxai model download --model_name=<your-model-name> --model_format="onnx"
  • python command:
from remyxai.api import download_model 

model_name = "<your-model-name>"
model_format = "onnx"
print(download_model(model_name, model_format))

Tasks

Train an image classifier:

  • cli command:
$ remyxai classify --model_name=<your-model-name> --labels="comma,separated,labels" --model_size=<int between 1-5>

add the optional --hf_dataset if you want to train with your own image dataset on 🤗. See the docs for more details

  • python command:
from remyxai.api import train_classifier

model_name = "<your-model-name>"
labels = ["comma", "separated", "labels"]
model_size = 3 # use 1 for microcontrollers

# Optional HF dataset
hf_dataset = "your/hf-dataset"

print(train_classifier(model_name, labels, model_size, hf_dataset))

Train an object detector:

  • cli command:
$ remyxai detect --model_name=<your-model-name> --labels="comma,separated,labels" --model_size=<int between 1-5>

add the optional --hf_dataset if you want to train with your own image dataset on 🤗. See the docs for more details

  • python command:
from remyxai.api import train_detector

model_name = "<your-model-name>"
labels = ["comma", "separated", "labels"]
model_size = 3

# Optional HF dataset
hf_dataset = "your/hf-dataset"
print(train_detector(model_name, labels, model_size, hf_dataset))

Train a text generator:

  • cli command:
$ remyxai generate --model_name=<your-model-name> --hf_dataset=<your/hf-dataset>

Your Huggingface dataset should have two columns with naming conventions like:

  • "question", "response"

  • "question", "answer"

  • "input", "output"

  • "prompt", "response"

  • python command:

from remyxai.api import train_generator

model_name = "<your-model-name>"
hf_dataset = "your/hf-dataset"

print(train_generator(model_name, hf_dataset))

Deploy

Launch a Triton Server containerized deployment for your model. Currently supported for generate models. More model types support coming soon!

System requirements

Please make sure you have Docker, Docker Compose, and the NVIDIA Container Toolkit are installed.

Deploy a model with:

  • cli command:
# Bring up
remyxai deploy --model_name="<your-model-name>"

# Bring down
remyxai deploy down --model_name="<your-model-name>"
  • python command:
from remyxai.api import deploy_model

model_name = "<your-model-name>"

deploy_model(model_name, action='up') # action can be "up" or "down"

And you can run inference with:

  • cli command:
remyxai infer --model_name="<your-model-name>" --prompt="Your prompt here"
  • python command:
from remyxai.api import run_inference

model_name = "<your-model-name>"
prompt="Your prompt here"

result, time_elapsed = run_inference(model_name, prompt, server_url="localhost:8000", model_version="1")
print(result)

User

Get user profile info:

  • cli command:
$ remyxai user profile
  • python command:
from remyxai.api import get_user_profile

print(get_user_profile())

Get user credit/subscription info:

  • cli command:
$ remyxai user credits
  • python command:
from remyxai.api import get_user_credits

print(get_user_credits())

Utils

Label images locally:

  • cli command:
$ remyxai utils label --labels="comma,separated,labels" --image_dir="/path/to/image/dir"
  • python command:
from remyxai.utils import labeler
model_name = "<your-model-name>"
labels = ["comma", "separated", "labels"]
image_dir = "/path/to/image/dir"
print(labeler(labels, image_dir, model_name))