Together: a cloud platform to build and run generative AI model training platforms

General Introduction

Together AI is a platform focused on generative AI models, providing a full range of services from model training and fine-tuning to inference. Users can utilize Together AI's highly efficient inference engine and GPU clusters to rapidly deploy and run various open source models. The platform supports multiple model architectures to meet different AI application requirements.

Large model chat demo interface

Function List

• Inference API: Supports reasoning on more than 100 open-source models and provides both serverless and dedicated instance models.
• fine tuning function: Allow users to fine-tune generative AI models using their own data, maintaining data ownership.
• GPU Cluster: Provides cutting-edge clusters of 16 to 1000+ GPUs to support large-scale model training.
• Custom Model Training: Train cutting-edge models from scratch and support multiple model architectures.
• multimodal model: Supports image recognition, image inference, image generation, etc.
• Efficient inference engine: Integrate the latest inference technologies such as FlashAttention-3 and Flash-Decoding to provide fast and accurate inference services.

Using Help

Installation and use

1. Register & Login::
   • Visit the Together AI website (https://www.together.ai/) and click the "Start Building" button to register.
   • Once registration is complete, log in to your account and go to the User Control Panel.
2. Select Service::
   • In the control panel, select the service module you need to use, such as the inference API, fine-tuning functions, or GPU clustering.
   • Select serverless or dedicated instance mode as required.
3. Inference API Usage::
   • Select the open source model you need to run, such as Llama-3, RedPajama, etc.
   • Integrate the model into your application through the API interface provided by Together AI.
   • Build your own RAG application using Together AI's embedded endpoints.
4. fine tuning function::
   • Upload your dataset and select the models that need to be fine-tuned.
   • Configure the fine-tuning parameters to start the fine-tuning process.
   • Once the fine-tuning is complete, download the fine-tuned model and deploy it.
5. GPU Cluster Usage::
   • Select the desired GPU cluster size and configure the hardware parameters.
   • Upload training data and model code to start training.
   • Once training is complete, the trained model is downloaded for inference or further fine-tuning.
6. Custom Model Training::
   • Select the Custom Model Training module to configure the model architecture and training parameters.
   • Upload the dataset and start the training process.
   • Once training is complete, the model is downloaded for deployment and inference.

workflow

1. Accessing the Control Panel: After logging in, go to the control panel and select the desired service module.
2. Configuration parameters: Configure inference, fine-tuning or training parameters as required.
3. Upload data: Upload the required dataset or model code.
4. Commencement of mission: Initiate reasoning, fine-tuning or training tasks and monitor task progress in real time.
5. Download results: After the task is completed, the model or inference results are downloaded for application integration.

common problems

• How to choose the right model?
  • Select the appropriate open source model based on the application scenario, such as text generation, image recognition, etc.
• What should I do if I encounter an error during the fine-tuning process?
  • Check the dataset format and parameter configurations and refer to the official documentation for adjustments.
• Poor performance during GPU cluster usage?
  • Confirm that the hardware configuration meets the requirements and adjust the training parameters to improve efficiency.

usable model

usage example

Example of model reasoning

import os
import requests

url = "https://api.together.xyz/v1/chat/completions"
payload = {
    "model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
    "max_tokens": 512,
    "temperature": 0.7,
    "top_p": 0.7,
    "top_k": 50,
    "repetition_penalty": 1
}
headers = {
    "accept": "application/json",
    "content-type": "application/json",
    "Authorization": "Bearer TOGETHER_API_KEY"
}

response = requests.post(url, json=payload, headers=headers)
print(response.text)

Example of model fine-tuning

import os
import requests

url = "https://api.together.xyz/v1/fine-tune"
payload = {
    "model": "togethercomputer/llama-2-70b-chat",
    "data": "path/to/your/data",
    "epochs": 3,
    "batch_size": 8
}
headers = {
    "accept": "application/json",
    "content-type": "application/json",
    "Authorization": "Bearer TOGETHER_API_KEY"
}

response = requests.post(url, json=payload, headers=headers)
print(response.text)