Sources
Loading...
Additional media
Loading...
DeepNewz, mobile.
People-sourced. AI-powered. Unbiased News.
You can now reproduce DeepSeek-R1's reasoning on your own local device! https://t.co/EppRWwI5Tt --- Newsletter https://t.co/lLfwtmvXkM More story https://t.co/yFb3Ds4tXm LinkedIn https://t.co/FC5hpfOlxr #AINewsClips #AI #ML #ArtificialIntelligence #MachineLearning https://t.co/ARUQkHmmVM
We managed to fit Llama 3.1 8B < 15GB with GRPO! Experience the R1 "aha moment" for free on Colab! Phi-4 14B also works with @UnslothAI & vLLM is now integrated allowing 20x faster inference! LoRA with GRPO also just work! 1. We removed double memory usage during vLLM serving…
You can now reproduce DeepSeek-R1's reasoning on your own local device! Experience the "Aha" moment with just 7GB VRAM. Unsloth reduces GRPO training memory use by 80%. 15GB VRAM can transform Llama-3.1 (8B) & Phi-4 (14B) into reasoning models. Blog: https://t.co/pjvgXOeHZQ https://t.co/10n7OBetkJ
Recent developments in AI model deployment have highlighted the capabilities of Llama 3.1 405B, recognized as one of the most demanding open-source models. It excels in multilingual dialogues and processing large datasets, offering exceptional performance and flexibility. Additionally, Llama 3.1 8B has been optimized to fit within 15GB of memory using GRPO, allowing users to experience the 'aha moment' for free on Google Colab. The integration of vLLM has reportedly enabled 20 times faster inference, while UnslothAI has reduced GRPO training memory usage by 80%. Furthermore, users can now reproduce the reasoning of DeepSeek-R1 on local devices with just 7GB of VRAM, transforming Llama-3.1 (8B) and Phi-4 (14B) into reasoning models.
