Vext v1.10: Media File Link as Data Source, Multi-File Upload, and More

Vext v1.10: Media File Link as Data Source, Multi-File Upload, and More

·

2 min read

We are excited to announce the release of Vext v1.10, which focuses on improvements to enhance your overall experience.

New Feature

You can now add custom public video links, YouTube, and Vimeo as data sources:

  • Custom Public Video Link: Video/audio file accessible directly via URL

  • Youtube: Any link but password protected or private ones

  • Vimeo: Any link but private ones

Vext will automatically transcribe the audio from these media links and convert it into a text file, ready for use in your vector database. This feature simplifies the process of incorporating rich media content into your projects.

Improvements:

Playground Redesign

The Playground has undergone a complete redesign, providing greater clarity on API structure. This new design makes it easier to understand what’s happening at the API level, simplifying the integration process and enhancing your development experience.

Note: The above requires you to:

  1. Set up input, learn more here.

  2. Set up a system prompt that takes in these custom variables.

Here's the Llama 3 70B system prompt that we've used for the above example:

<|begin_of_text|><|start_header_id|>system<|end_header_id|>
You are a helpful assistant, assist users with their questions. 
He is in ${user_location}, his name is ${user_name}.

<|eot_id|> <|start_header_id|>user<|end_header_id|> 
${payload} 
<|eot_id|> 
<|start_header_id|>assistant<|end_header_id|>

Multi-File Upload

We've expanded the multi-file upload capability. You can now upload up to 5 files, each up to 10MB, per upload session when adding files to your dataset. This improvement significantly boosts your efficiency when working with large datasets.

Bye-bye "Query", hello "Credit"

We’ve made changes to our terminology to improve clarity. Previously, we counted "queries" as the basic unit when an LLM is triggered. Based on user feedback, we are now renaming this unit to "credit."

Additionally, we’re introducing a new credit calculation method. Different LLMs will now consume different amounts of credit, with basic models counting as 1 credit, while more advanced models like OpenAI GPT-4 and Claude 2 will consume more credits. For more details, please refer to our new credit calculation method here.