Hugging Face has had a busy week filled with AI advancements, launching Transformers Agent, a new tool that enables users to manage over 100,000 HF models through conversation with the Transformers and Diffusers interface. This latest API suite can generate other HF models to solve complex, multimodal problems. Transformers Agent offers a natural language API, an array of curated tools, and an agent capable of interpreting natural language and utilizing these tools. This extensible system can integrate any relevant community-developed tools. Users must initiate an agent, a Long Language Model (LLM), before using the agent.run function. The system supports both OpenAI modes and open-source alternatives from BigCode and OpenAssistant. The tools are single functions with designated names and descriptions, each designed to perform a specific task. The agent learns how to utilize these tools through prompts that demonstrate how they can achieve the desired outcome. Users have two APIs at their disposal: single execution (run) and chat-based execution (chat). The run method enables the agent to select and use the necessary tools for the task, while the chat method is useful for maintaining the state across different instructions. The user-provided code runs using a small Python interpreter. To mitigate arbitrary code execution risks, only functions provided by Hugging Face and the print function can be executed, and attribute lookups and imports are not allowed. The curated tools range from document-based question answering using Donut and text-based question answering with Flan-T5, to image segmentation with CLIPSeg and text summarization with BART. Custom tools include a text downloader, a text-to-image creator, an image transformer, and a text-to-video generator. Hugging Face has also recently partnered with ServiceNow to create a new open-source language model for codes named StarCoder. Part of the BigCode initiative, StarCoder has surpassed larger models like PaLM, LaMDA, and LLaMA in performance and has demonstrated comparability or superiority to closed models like OpenAI’s code-Cushman-001. ------------------------- Get our Daily AI Newsletter: ⁠https://AIBox.ai⁠ Join our ChatGPT Community: ⁠https://www.facebook.com/groups/739308654562189/⁠ Follow me on Twitter: ⁠https://twitter.com/jaeden_ai⁠ See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related