Thanks Jason for sharing!

#CogVLM and #CogAgent ! github.com/THUDM/CogVLM

I am quite excited about this work! Ideation and hypothesis generation are challenging tasks for scientists. It is quite plausible that LLMs will contribute to the next breakthrough in science.

We crossed 1M models on Hugging Face!

An exciting lineup! Our #ChatGLM founder will be there too jietang see you soon.

🌟Dive into the future of AI this summer! #Tsinghua DCST presents a comprehensive summer school on #LLM . Master the art of pre-training, fine-tuning, and beyond. 🚀Open to all 2nd & 3rd year CS undergrads. 🗓Apply before May 1st! 💡Learn more at ss.cs.tsinghua.edu.cn

GitHub Copilot and ChatGPT competitor from #China that you probably don't know exist.

1. General Language Model - GLM by Tsinghua KEG (THUDM) & ZhipuAI has new improved GLM-4 model. Unlike OpenAI they have open source model.

GLM-4 explaining components of model architecture from image👇

OAG-Challenge at KDD Cup 2024 has launched! Welcome to join the competition and address the key challenges in academic graph mining. The use of LLMs is encouraged and a free quota of GLM-4 API is provided.

AutoWebGLM

Bootstrap And Reinforce A Large Language Model-based Web Navigating Agent

Large language models (LLMs) have fueled many intelligent agent tasks, such as web navigation -- but most existing agents perform far from satisfying in real-world webpages due to three

ChatGLM-Math

Improving Math Problem-Solving in Large Language Models with a Self-Critique Pipeline

Large language models (LLMs) have shown excellent mastering of human language, but still struggle in real-world applications that require mathematical problem-solving. While many

ChatGLM-Math: Improving Math Problem-Solving in Large Language Models with a Self-Critique Pipeline

Achieves SotA results for opensource LLMs on GSM8K and MATH by self-critique

arxiv.org/abs/2404.02893

🏆 Another exciting news! We announce the second KDD Cup 2024 challenge!
Meta KDD Cup 2024 - CRAG: Comprehensive RAG Benchmark
Let the games begin! 🌟🏃‍♂️

A huge thanks to Meta for making this happen!

aicrowd.com/challenges/met…

CogView3: Finer and Faster Text-to-Image Generation via Relay Diffusion

abs: arxiv.org/abs/2403.05121

Introduces CogView3, which uses relay diffusion (a variant of cascaded diffusion) in latent space with a 3B U-net and T5 XXL text encoder. Trained with LAION-2B, recaptioned…

CogView3

Finer and Faster Text-to-Image Generation via Relay Diffusion

Recent advancements in text-to-image generative systems have been largely driven by diffusion models. However, single-stage text-to-image diffusion models still face challenges, in terms of

CogView3: Finer and Faster Text-to-Image Generation via Relay Diffusion

The distilled variant of CogView3 achieves comparable performance while only utilizing 1/10 of the inference time by SDXL

arxiv.org/abs/2403.05121

Jan Schnyder AK Hugging Face Allen Institute for AI UCSB NLP Group Waterloo's Cheriton School of Computer Science CogVLM added! Please have a try! :D 🚀

Bill Yuchen Lin 🤖 AK Hugging Face Allen Institute for AI UCSB NLP Group Waterloo's Cheriton School of Computer Science This is awesome, any plans on adding CogVLM? It outperforms most other models on our benchmarks.

Bill Yuchen Lin 🤖 AK Hugging Face Allen Institute for AI UCSB NLP Group Waterloo's Cheriton School of Computer Science Sick!! CogAgent might also be interesting to add (huggingface.co/THUDM/cogagent…). Its an improved version based on CogVLM by Tsinghua KEG (THUDM) . IMO currently the best open-source MLLM one can get, the people at THUDM are cooking 🧑‍🍳

🏅 To me, this feels more like the kind of neural model interpretability research we should be doing than much of the recent work on interpretability of transformer models.

LongAlign - A recipe for long context alignment of LLM introduced by Tsinghua KEG (THUDM) 🔥

✨ With LongAlign-10k dataset to support
✨ Outperforms existing LLM recipes by up to 30%, while maintaining the ability to handle short, general tasks

Model huggingface.co/THUDM/LongAlig…
Paper…