Before you select a foundational model, make sure you have collected and preprocessed a specialized dataset relevant to your use case
Currently Supported Models (More coming soon)
Currently Supported Models (More coming soon)
| Model Name | Developer | Model Size | Description |
|---|---|---|---|
| GPT-4 | OpenAI | 800 GB | High performance, versatile model for advanced chatbot applications |
| GPT-4-1106-preview | OpenAI | 1.2 TB | Latest version with cutting-edge features and improvements |
| GPT-4-turbo | OpenAI | 1.5 TB | Fastest response times, optimized for speed and efficiency |
| GPT-4o | OpenAI | 1.1 TB | Optimized for specific tasks |
| GPT-3.5-turbo | OpenAI | 350 GB | Cost-effective, fast responses, suitable for general-purpose chatbots |
| RedPajama-INCITE-7B-Chat | TogetherAI | 1.12 GB | Lightweight, efficient model designed for chat applications |
| Llama-2-7B-32K-Instruct | Meta, TogetherAI | 308 MB | Handles extended context, ideal for long conversations |
| Mistral-7B-Instruct-v0.1 | MistralAI | 6.85 GB | High accuracy, instruction-tuned for better task following |
| Mistral-7B-Instruct-v0.2 | MistralAI | 5.31 GB | Enhanced instruction-following capabilities for improved interaction quality |
| OpenHermes-2p5-Mistral-7B | MistralAI | 166 MB | Hybrid model, excels in handling complex and diverse tasks |
| Nous-Hermes-2-Mixtral-8x7B-DPO | Mixtral | 662 MB | Optimized diverse parameters for versatile chatbot applications |
| Mixtral-8x7B-Instruct-v0.1 | Mixtral | 166 MB | Instruction-tuned, efficient for specific tasks with high precision |
| Nous-Hermes-2-Mixtral-8x7B-SFT | Mixtral | 166 MB | Supervised fine-tuning ensures high precision and accuracy in responses |
Factors to Consider
Cost
Cost is a major factor, especially for large-scale deployments. Models like GPT-4-turbo and GPT-3.5-turbo offer cost-effective options without significantly compromising on performance. Generally, more advanced models with higher capabilities tend to be more expensive. Ensure that the cost aligns with your budget and expected usage volume.Model Size
Model size impacts both the hardware requirements and the inference speed. Larger models like GPT-4 and GPT-4-1106-preview require substantial computational resources but offer higher accuracy and more sophisticated capabilities. Smaller models, such as Llama-2-7B-32K-Instruct, are more lightweight and can be deployed on less powerful hardware.As hardware performance accelerates, model size becomes less of a cost and speed burden.
Ease of Fine-tuning
Fine-tuning allows you to adapt the model to your specific use case, improving accuracy and relevance. Models with robust fine-tuning capabilities, like Mistral-7B-Instruct, are preferable for applications that require specialized knowledge or behavior.Specific Use Cases
Certain models are optimized for specific applications such as chatbots, long-context handling, or diverse task execution. Understanding your primary use case can guide you to the most suitable model.Evaluating models for a production environment?
We’re here to help. Contact us to help walk you through the testing, evaluation, and benchmarking processes
for your use case.