AI Briefing#huggingface
The latest AI and developer news about #huggingface, with the original source and a short summary.
The latest AI and developer news about #huggingface, with the original source and a short summary.
HuggingFace has launched TTS Arena, which compares the quality of TTS models based on human preferences.
Hugging Face has released a guide on using PEFT and QLoRA to efficiently fine-tune Gemma models.
Google has released Gemma, an open LLM family built on Gemini technology.
Hugging Face's PEFT library now supports new merging methods that can efficiently combine LoRA adapters.
This presents a method for building low-cost, high-efficiency custom models that achieve GPT-4-level performance by leveraging synthetic data generated by LLMs.
AMD and Hugging Face are hosting an AI developer contest with a total prize pool of $160,000.
Hugging Face has launched a Messages API compatible with the OpenAI API to support transitions to open LLMs.
Qwen1.5, the next-generation model in the Qwen series, has been released with various sizes and improved performance.
SegMoE, a Mixture-of-Experts (MoE) framework for diffusion models, has been released.
This explains the structure of PatchTST, a model that improves time-series forecasting performance, and how to use it with the Hugging Face library.
Hugging Face's LLM serving solution TGI is now officially supported on AWS Inferentia2 and Amazon SageMaker.
HuggingFace has released an implementation methodology, related tools, and datasets for **Constitutional AI** for open source models.
Patronus AI has launched a new LLM leaderboard that evaluates real-world enterprise use cases such as finance and law.
An open-source leaderboard for evaluating factuality and faithfulness hallucinations in LLMs has been released.
Hugging Face and Google Cloud have entered into a strategic partnership to support building AI based on open models.
Open-source LLMs like Mixtral have reached a level where they can perform ReAct-style agent workflows.
This covers how to use Meta's Wav2Vec2-BERT to improve automatic speech recognition (ASR) performance for languages with limited data.
It introduces the technical method of building a custom LLM leaderboard using Hugging Face's new template, along with Vectara's case study.
The Unsloth library can speed up LLM fine-tuning by up to 2x while reducing memory usage by 40%.
aMUSEd, an efficient non-diffusion text-to-image generation model that recreates Google's MUSE, has been released.
By applying Speculative Decoding, the inference speed of Whisper was doubled while maintaining accuracy.
This summarizes the development process of open source LLMs in 2023 and their core technical components.
This explains in detail how the MoE architecture works, how it is trained, and the trade-offs during inference.
Mistral AI's new MoE model Mixtral 8x7b has been released, and it outperforms GPT-3.5.