Prezi, a visual communications software company, has joined the Hugging Face Expert Support Program to leverage modern machine learning's potential. The company has been integrating smaller, more efficient open-source models into their ML workflows. This cooperation started at a perfect time, as multimodal models are becoming increasingly capable.
AI | Rating: 58 | 2024-06-25 02:05:48 PM |
XLSCOUT, a Toronto-based company, has developed a powerful proprietary embedding model called ParaEmbed 2.0 in collaboration with Hugging Face's Expert Support Program. The model is tailored for patents and IP, enhancing the understanding and analysis of complex patent documents. This technology allows XLSCOUT's products to offer better performance for drafting patent applications, patent invalidation searches, and ensuring novel ideas.
AI | Rating: 83 | 2024-06-25 11:56:28 AM |
Reddit announced a partnership with Google to power a new Generative AI-based search engine using Retrieval Augmented Generation (RAG). However, the attempt did not go as planned, resulting in incorrect recommendations. High-quality data is crucial for AI systems to produce high-quality outputs, and prioritizing data quality is essential from the outset.
AI | Rating: 81 | 2024-06-24 06:36:01 PM |
Microsoft released Florence-2, a cutting-edge vision-language model, in June 2024. The model is small in size, with 0.2B and 0.7B parameters, and performs well on various computer vision and vision-language tasks. It supports tasks such as captioning, object detection, and OCR out of the box. However, users may need to fine-tune the model for specific tasks or domains.
AI | Rating: 85 | 2024-06-24 03:35:56 PM |
Hugging Face and Argilla collaborated on the Data Is Better Together initiative, aiming to empower the open-source community to create impactful datasets collectively. The project focused on the prompt ranking project, creating a dataset of 10K prompts, both synthetic and human-generated, ranked by quality. The initiative has been organized into two sections: community efforts and cookbook efforts, allowing everyone to contribute.
AI | Rating: 81 | 2024-06-22 03:39:06 AM |
BigCodeBench is a benchmark for evaluating large language models on solving practical and challenging programming tasks. The current benchmark, HumanEval, is criticized for being too simple and not representative of real-world programming tasks. The new benchmark aims to address these concerns by including algorithm-oriented tasks and diverse libraries and function calls.
AI | Rating: 85 | 2024-06-17 02:15:30 PM |
Hugging Face Accelerate exposes two popular implementations of the ZeRO Redundancy Optimizer algorithm, one from DeepSpeed and the other from PyTorch. The organization upstreamed a precision-related change and a concept guide to enable seamless switching between the backends. A training pipeline was run with DeepSpeed and PyTorch FSDP, resulting in differing results.
AI | Rating: 88 | 2024-06-13 06:35:32 PM |
Stability AI has released Stable Diffusion 3 (SD3), a latent diffusion model with 2B parameters, available on the Hugging Face Hub and compatible with Diffusers. SD3 consists of three text encoders, a Multimodal Diffusion Transformer, and an AutoEncoder model. The model is now available for use with Diffusers.
AI | Rating: 62 | 2024-06-12 08:25:08 PM |
The RLOO (REINFORCE Leave One-Out) Trainer is a new online RLHF training algorithm designed to be more accessible and easier to implement. It requires less GPU memory and takes less wall time to converge. RLOO uses approximately 50-70% less vRAM than PPO, runs 2x faster than PPO with 1B models, and up to 3x faster than PPO with 6.9B models.
AI | Rating: 85 | 2024-06-12 02:06:12 PM |
The author joined Hugging Face nearly three years ago and noticed significant changes in the Transformers documentation. The documentation initially focused on text models for natural language tasks, but expanded to include new models and usage patterns as transformer models became the default approach to AI. The documentation was updated incrementally without considering the evolving audience and library.
AI | Rating: 81 | 2024-06-07 04:35:15 PM |