Recent developments in the field of natural language processing (NLP) and machine learning (ML) have seen the introduction of new models and updates. LLaVa-NeXT, also known as LLaVa-1.6, has been integrated into Hugging Face Transformers, surpassing its predecessor LLaVa-1.5 with enhancements like higher image resolutions, improved OCR and reasoning capabilities, and the inclusion of various LLMs such as Mistral-7B and Yi-34B by NousResearch. Additionally, the latest MLX LM features advancements like (Q)LoRA fine-tuning compatibility with OpenAI format chat and completions data, as well as the ability to fuse and export LoRA fine-tuned models to GGUF in *fp16 only* format. The focus on open-source LLM models is increasing, aiming to elevate the standard of such models to match proprietary LLMs in human evaluation.
Free LLM Improvements? 🤔 Model Merging allows us to blend/stack multiple open LLMs into one—bigger or the same size—without extra training to extend skills and performance!🌱 @arcee_ai just released the paper for their open-source library mergekit. Let's take a look 👀 How to… https://t.co/TRZiwF2e36
Transformers 4.39 is out, and it's packed with exciting updates! 🚀 New models: Mamba, Command-R, LLaVA-NeXT, MusicGen Melody, StarCoder2, SegGPT, ... ⚡️GaLore optimizer for accessible pre-training 🤏Quanto integration and Exllama+AWQ 🍎MLX support https://t.co/AsqciYOi2e
Super excited for our next open-source LLM model drop!! We will be pushing up the SOTA for open-source models significantly. This time, the focus is purely on usable LLMs that match proprietary LLMs in human eval. i.e., the benchmark that really matters... https://t.co/gYBDUt5qRx
Latest MLX LM has a couple of nice additions: - (Q)LoRA fine-tuning works with OpenAI format chat and completions data (h/t @Madroidmaq) - Fuse and export LoRA fine-tuned models to GGUF *fp16 only* (h/t @LiMzba) pip install -U mlx-lm
Excited to share that LLaVa-NeXT (also known as LLaVa-1.6) is now in @huggingface Transformers! Improves upon its predecessor, LLaVa-1.5, by incorporating * higher image resolutions * better OCR and reasoning capabilities * various LLMs (Mistral-7B, Yi-34B by @NousResearch) 1/2 https://t.co/BBpUnnfcTg