Hugging Face has released Transformers version 4.42, which includes new models like Gemma 2, RT-DETR, InstructBlip, and LLaVa-NeXT-Video. Gemma 2 models, developed by Google, show superior performance in language understanding tasks. RT-DETR is a real-time object detection model based on transformer architecture. InstructBlip enhances visual instruction tuning, and LLaVa-NeXT-Video combines video and image datasets for video understanding tasks.
The release also includes tool usage improvements, RAG support, GGUF fine-tuning, and quantization enhancements like a quantized KV cache. New instance segmentation examples have been added, and deprecated components have been removed. These updates solidify Hugging Face’s position as a leader in NLP and machine learning.
The release also introduces JSON schema descriptions for Python functions, a standardized API for tool models, and support for GGUF fine-tuning. The quantization improvements reduce memory requirements for generative models, and new instance segmentation examples have been added.
Overall, Transformers 4.42 is a significant development for Hugging Face, offering new models, enhanced tool support, and various optimizations. The release cements Hugging Face’s position as a leader in NLP and machine learning.
Source link
Source link: https://www.marktechpost.com/2024/06/29/transformers-4-42-by-hugging-face-unleashing-gemma-2-rt-detr-instructblip-llava-next-video-enhanced-tool-usage-rag-support-gguf-fine-tuning-and-quantized-kv-cache/?amp