Woooh finally finally finally it is here!!!!!!! Qwen2-VL! 2B and 7B openweight under Apache 2.0 and 72B for API temporarily! Check the SOTA performance on vision understanding benchmarks and check the new capabilities of understanding videos and playing as a visual agent! Have

Qwen 2VL 7B & 2B are here - Apache 2.0 licensed smol Vision Language Models competitive with GPT 4o mini - w/ video understanding, function calling and more! 🔥 > 72B (to be released later) beats 3.5 Sonnet & GPT 4o > Can understand up to 20 min of video > Handles arbitrary

Thx King 🎉

Some notes. We failed at supporting llama.cpp (damn it is a bit too hard but we’ll try again) and thus temporarily no ollama or lmstudio for now

Does style matter over substance in Arena? Can models "game" human preference through lengthy and well-formatted responses? Today, we're launching style control in our regression model for Chatbot Arena — our first step in separating the impact of style from substance in

Just made a Space on Hugging Face to demo the great Qwen2-VL-2B model just released by Qwen! 🔥 This thing is fast!!! Thanks to ZeroGPU and the fact that this model is only 2B in size! 🚀 PS: merve your Baklava test has passed! 🤗

Talk to this man for the API issues!

Always fun to be on Thursdai!

Super cool tool for me to explain to newbies about the requirements of your computer resources!

The new Qwen2-VL-7B Instruct model gets *100%* accuracy extracting text from this handwritten document. This is the first open weights model (Apache 2.0) that I've seen OCR this accurately. (Thank you Florent Daudens for the tip!) huggingface.co/spaces/Ganymed…

We are still alive... GitHub flagged our org for unknown reasons and we are trying to approach them for some solutions.

I just don't know what happened when we guys are still working on GitHub but last night suddenly QwenLM github org was flagged and since then contents have not been shown in the public. I advise you to turn to huggingface or our discord for contact temporarily.

Thx for the appreciation☺️

Guys, we are still suffering from the problem, and feel free to turn to HF / ModelScope or our Discord server for discussion!

We are fucking back!!! Go visit our github now!

this is an official Qwen-2 VL stan account

Back now

💗💗💗

Cong

I personally think the key factor may not be the "reflection" technique itself, but rather the additional tokens it forces the model to generate. This concept could be connected to [PAUSE] tokens, as described in arxiv.org/pdf/2310.02226. By generating more tokens, the model