Infinite.Tech Mistral API Integration & Roadmap

Introduction

Mistral AI, a promising breakout AI startup based in France, has recently released an impressive set of API and open local model tools that have caught the tech world's attention. Infinite.Tech, always on the lookout for powerful, accessible AIs, is excited to incorporate these tools into our platform so you can test and explore their capabilities to find how to use them in the way that suits you. With a reputation and proclamation to high scientific standards with the open tech championing aligning with French Revolutionary ideology, this team and models are proving to be a force in AI and LLM access.

Check it out today on the Infinite Tech App and try what the Mistral chat completion docs suggest: starting a conversation about French cuisine to express their palate of the fine, pungent sophistication of gooey cheeses.

Signing Up and Connecting Keys With Infinite.Tech

First, sign up for a beta account to use Mistral AI's API. Once you have an account, you can add your API keys to the Infinite.Tech App and use your account.

1. Login or sign up to the Mistral Console

Mistral AI’s approach blends advanced AI capabilities with openness and community involvement, making it a versatile tool for developers and researchers. You can visit their API documentation and leading website for a detailed understanding of their API and models.

1.5. Set Billing Limits

Always a good idea

Building this feature, testing, and tutorials have taken about 276,844 tokens, which cost 0.97 EUR.

Mistral AI provides an API with pay-as-you-go access to its latest models. This API is currently in beta, allowing users to test and utilize the chat and embeddings endpoints. Examples of API usage include generating chat completions and embeddings for specified inputs.

2. Navigate to the API Keys tab and Generate a new API key

Mistral AI emphasizes open science and free software, releasing many models and tools under permissive licenses. They encourage community contributions and offer transparent access to their weights for full customization.

3. Copy And Paste the Generated Key into your App.Infinite.Tech Profile

Click on your user profile in the right and set the key in the profile

4. Use the Mistral Model Catalog

Generating Chat Completions: The API allows users to send prompts for generating chat completions, with control over parameters like temperature and max tokens.

First Impressions

The models respond with information-rich answers and transform data with rational, accurate thinking across modes. The tiny model is super quick, and we expect gains with local hardware hosting, which is on our roadmap.

It is particularly important with these tools to pick the correct model, adjust the settings, and have clear prompts with instructions to avoid repetitiveness and achieve expected or optimal results.

Open Models and Weights

Mistral is setting a standard for open model access and is aiming us toward integrating more open models and hosing protocols. These models are available under the Apache 2.0 License. They can be accessed on platforms like Hugging Face or from Mistral AI’s documentation. This open-source approach fosters community contribution and enhances transparency and use potential. Mixtral 8x7B Model: A key offering of Mistral AI is the Mixtral 8x7B model, known for its speed and power. It excels in various benchmarks, supports multiple languages, and has coding capabilities with a 32k sequence length capacity. The model is available for use through the Mistral API or self-deployment.

Initial Experiments

To better understand Mistral AI’s capabilities, Infinite Tech conducted several experiments. In cheese, as suggested, cross-language understanding, gardening, and the Louvre museum. These tests suggest that Mistral AI has a high quality of multi-mode information thinking and might have an edge in understanding translations across multiple languages compared to primarily one-language trained models.

Fromager-Affineur

Given the same prompt, “What do you have to say about French Cheeses.” It seems to give longer and more concrete answers than GPT-4. This trait seems repeatable as the models seem talkative and continue freely.

Multi-Language Tutor

Given the model's bi-lingual origin and expertise in this area, it would be interesting to see how it performed as a language tutor. I asked questions on how language structure differed and then transformed them into lessons using the quiz module.

Roadmap

1. BETA API Keys → [x]

2. BETA API Model Inferencing → [x]

3. BETA API Embeddings →[ ]

4. Local Host Inferencing→[ ]

Closing

Visit Infinite.Tech and explore these new models yourself. Get intuition in their collection of AI models and appreciate the exquisite taste it brings.

You can visit their API documentation and main website for a more detailed understanding of Mistral AI’s API and models.