Together AI: AI Acceleration Cloud for Fast Inference and Training

[baoshi.rao]

Introduction

Together AI is an AI Acceleration Cloud designed to streamline the generative AI lifecycle. It provides a comprehensive platform for fast inference, fine-tuning, and training of AI models, leveraging scalable GPU infrastructure and easy-to-use APIs.

What is Together AI?

Together AI is an end-to-end platform for generative AI, supporting over 200 models across various modalities like chat, images, code, and more. It offers OpenAI-compatible APIs and scalable infrastructure for running, fine-tuning, and training models.

How to Use Together AI

Users can interact with Together AI through:

• Serverless Inference APIs for quick model deployment.
• Dedicated Endpoints for custom hardware deployment.
• Fine-Tuning via simple commands or API-controlled hyperparameters.
• GPU Clusters for large-scale training.

The platform also provides a web UI, API, and CLI for managing endpoints and services.

Core Features

• Serverless Inference API: For open-source models.
• Dedicated Endpoints: Custom hardware deployment.
• Fine-Tuning: Supports LoRA and full fine-tuning.
• Together Chat App: For open-source AI interactions.
• Code Sandbox & Interpreter: For AI development environments.
• GPU Clusters: With NVIDIA GPUs (GB200, B200, H200, H100, A100).
• Extensive Model Library: 200+ generative AI models.
• OpenAI-Compatible APIs: For seamless integration.
• Accelerated Software Stack: Includes FlashAttention-3 and custom CUDA kernels.
• High-Speed Interconnects: InfiniBand and NVLink.
• Management Tools: Slurm and Kubernetes for robust operations.

Use Cases

1. Enterprise AI Model Training: Used by companies like Salesforce and Zoom.
2. AI Customer Support Bots: Scalable solutions for high message volumes.
3. Production-Grade AI Apps: Unlocking data for developers and businesses.
4. Text-to-Video Models: Used by platforms like Pika.
5. Cybersecurity Models: Developed by companies like Nexusflow.
6. Custom Generative AI Models: Built from scratch.
7. Multi-Document Analysis: For complex data tasks.
8. Code Generation & Debugging: With advanced LLMs.
9. Visual Tasks: Advanced reasoning and video understanding.
10. Data Classification & Extraction: Structured data processing.

Pricing

• Serverless Inference: Varies by model and token count.
• Dedicated Endpoints: Customizable GPU endpoints with per-minute billing.
• Fine-Tuning: Based on model size, dataset size, and epochs.
• GPU Clusters: Starting at $1.30/hour.
• Code Execution: Per hour or per session.

For detailed pricing, visit Together AI Pricing.

FAQ

• Supported AI Models: Over 200 generative models across various modalities.
• GPU Hardware: Includes NVIDIA GB200, B200, H200, H100, and A100.
• Performance Optimization: Through accelerated software and high-speed interconnects.
• Fine-Tuning Capabilities: Yes, via LoRA or full fine-tuning.
• Enterprise Suitability: Designed for enterprise-scale AI operations.

Contact

For more information, visit Together AI Contact.

Company

Together AI is based in San Francisco, CA. Learn more at Together AI About.