Computer

Phind grows: chatbot, search engine and Visual Studio Code assistant thanks to AI

Phind grows: chatbot, search engine and Visual Studio Code assistant thanks to AI

In November 2023 we presented Pindan AI-based generative model that even then surpassed OpenAI GPT-4 in creating programming code.

Today comes the news of the release of PHIND-70Ban advanced language model that gives its best in generating code and solving technical problems but which, obviously, can be used for general purposes.

The main feature of Phind-70B is the execution speed: Processing up to 80 tokens per second, the model comes across as four times faster than GPT-4 Turbo. “Behind the Scenes,” is based on CodeLlama-70B, but was further refined by adding an additional 50 billion tokens. The developers used the library TensorRT-LLM of NVidia on H100 GPU, pointing out that there is room for further optimizations useful for increasing the performance of Phind-70B.

Phind-70B supports a context window of size equal to 32,000 tokens. With the expression context window, refers to the amount of information that a language model takes into account to generate an answer or make a hypothesis at a given time. Such a large context window (32,000 tokens) allows the model to consider more past information, which can support the generation of more accurate and consistent responses.

The model is already available to be tried for free without the need for registration. By subscribing to a subscription plan Phind Prousers can remove all limits of use.

In the Phind-70B presentation note, the company thanks its cloud partners SF Compute and AWS, along with Meta and NVidia, for their support in implementing and delivering the model. As a side note, he concludes by revealing that he burned an H100 GPU during Phind-70B training, confirming the computational effort of the process.

Phind-70B AI model

How to use Phind

You can test your Phind skills directly from your web browser, from both desktop and notebook systems and mobile devices. By connecting to the official project website, you will find two tabs: Search e Chat. The first activates the web search and the processing of the results obtained with artificial intelligence.

In other words, Phind Search allows us to best respond to the questions raised by the user by combining the results coming from the contents of the web pages deemed most reliable and reliable from time to time.

By clicking on instead Chat at the top, the chatbot that is, the mechanism that is based solely on the generative model and the LLM (Large Language Model) previously trained to provide useful and relevant answers. Furthermore, the chatbot is multi-channel (it supports the use of images to support the text produced in output) and supports the launch of actual conversations developed on continuous “question and response”.

With a click on the appropriate drop-down menu, it is possible to opt for the use of the smaller model, PHIND-34Bwhich currently has no limits in terms of use.

Michael Royzenco-founder and CEO of Phind with previous experience in Microsoft, Cloudflare, SmartLens and Lyft, points out that his project concretely looks at the world open source. Phind engineers, in fact, have committed to publicly releasing the weights of the Phind-34B model over the next few weeks. Subsequently, the same will be done in the case of Phind-70B as well.

Shortcuts to use Phind faster

Given that Phind can also be used from Android and iOS mobile devices without installing any dedicated app but by referring to the native website, to interact with the model even more quickly some useful shortcuts from keyboard:

  • / – Allows access to the search box
  • Cmd/Ctrl+K – Start a new search
  • ESC – Stops the response generation procedure
  • Cmd/Ctrl+Enter – by positioning yourself on the search box, send the query to Phind without carrying out a web search

Connect Phind with Visual Studio Code to generate programming code

We said that Phind offers excellent performance in generating programming code. A bit like you can do with others toolfor example GitHub Copilot, you can add theextension official for Visual Studio Code which allows you to directly manage the development of any application in the Microsoft editor.

Just access the page Phind, chat with your codebaseclick on Installallow Visual Studio Code to open (which must already be installed on your system) and then select the button again Install and finally Trust Workspace & Install.

Phind Visual Studio Code integration

Per activate the extension, you need to click on the Phind icon in the left column of Visual Studio Code and then log in “one-off” with a Google or GitHub account. With one click Activate Extension finally you must authorize the start of the Microsoft editor (Apri Visual Studio Code).

Generate programming code with Phind and VS Code

Provided the appropriate permissions, Phind shows all the instructions for communicate with the chatbot integrated into Visual Studio Code. The symbol @for example, allows you to specify specific files that make up the project while with CTRL+MAIUSC+I you can select a block of code and pass it to Phind to obtain analysis, understand how to improve it, solve a problem or bug that cannot be identified.

Create programming code and correct it with Phind

By pressing the key combination CTRL+MAIUSC+M you can generate code in-line. This means that, using simple or complex commands in natural language, also using the Europen language, you can ask Phind to generate code in any programming language.

Generate code inline within Visual Studio Code

By clicking on the icons that highlight errors or warnings, during the development activity, you can ask Phind for intervention or in any case a solution consultancy. In this way, thanks to artificial intelligence, the programming work can proceed quickly, overcoming those obstacles which in some cases can be truly blocking.

Opening image credit: iStock.com – Userba011d64_201

Leave a Reply

Your email address will not be published. Required fields are marked *