Скачать или смотреть Dots.LLM1 by Rednote — In-Depth Testing (A VERY Fun MoE Model)

Timestamps:

00:00 - Intro
00:48 - First Look
02:38 - Technical Look
07:08 - Python Game Test
09:07 - HTML Site Test
11:51 - SaaS MVP Refusal Test
13:48 - Johnny’s Duplicator Test
15:38 - JS Retro Web Game Test
17:38 - Closing Thoughts

Join the AI Discussions Discord:   / discord  
AI Integration Consulting: https://bijanbowen.com

In this video, we take an in-depth look at Dots.LLM1, a newly released 142B MoE model from Rednote HiLab. According to its creators, this model was trained without using synthetic data during pretraining — a notable distinction compared to many recent frontier models. It’s also Rednote’s first major open-source model release, and it comes alongside a set of custom training and efficiency innovations developed during its creation.

We test the model across a range of creative and coding tasks, including Python game generation, HTML website building, and JavaScript retro game development. The model also shows potential in handling humor, refusals, and generating structured MVP-style output. Despite its size, Dots.LLM1 is surprisingly responsive and expressive, offering a unique personality for a model in this class.

Model on Hugging Face:
https://huggingface.co/rednote-hilab/...