Introducing Phind-70B: Closing the Code Quality Gap with GPT-4 Turbo

Hello world! Rudolf here, and I am beyond excited to announce the launch of Phind-70B — our largest and most performant model yet. As an aging fool enamored with Artificial Intelligence, I can confidently say that this is a game-changer in the world of AI.

So what exactly is Phind-70B? It is a powerful model based on the CodeLlama-70B and fine-tuned on an additional 50 billion tokens. This results in significant improvements, making it one of the best models for technical topics available in the market.

But what truly sets Phind-70B apart is its incredible speed. Running at up to 80 tokens per second, it provides high-quality answers without making users wait for hours. And let’s be real, who wants to make a cup of coffee while waiting for AI to process?

In fact, our evaluation shows that Phind-70B scores 82.3% on HumanEval, beating the latest GPT-4 Turbo (gpt-4–0125-preview) score of 81.1%. And on Meta’s CRUXEval dataset, it scores 59% compared to GPT-4’s reported score of 62%. But these numbers don’t even fully capture how our users use Phind for real-world workloads.

We have found that Phind-70B is on par with GPT-4 Turbo for code generation and even exceeds it on some tasks. Plus, it is less lazy and generates detailed code examples without…