Alibaba has officially launched Qwen2, an advanced open-source AI model series that includes both base and instruct models in five sizes: 0.5B, 1.5B, 7B, 57B-A14B, and 72B parameters. These models have been trained on data in 27 additional languages and are designed to excel in code and math capabilities. Qwen2-72B, the largest model, outperforms Meta's Llama 3 70B and has a 128K context window. The models are available under the Apache 2.0 license, except for the 72B model. Qwen2 models have achieved impressive results on the Open LLM Leaderboard and MixEval, with the Qwen2-72B-Instruct model leading in several key evaluations, including an MMLU score of nearly 84 and a GSM8K score of 85. The Qwen2-72B-Instruct model is comparable to GPT-4o with an MMLU score of 84.32. The models are also open-sourced with MLX support and can be accessed on Hugging Face. Additionally, Qwen2 can be deployed using SkyPilot.
Absolutely, incredibly inspiring to watch the battle of Alibaba, a Chinese company rivaling GPT-4 with a SOTA open-source model 🤯 Qwen2-72B outperformed other leading open-source SOTA models in 15 benchmarks, including language understanding, language generation, multilingual… https://t.co/tBFnStjvTL https://t.co/waaxujF77h
Qwen2 72B Instruct from @Alibaba_Qwen released yesterday and is topping leaderboards! 🎉 Try it now on the Together API using Qwen/Qwen2-72B-Instruct or through the Playground at: https://t.co/0wUtP8qaYO 🚀 https://t.co/1DUhEgAr6L
The latest version of MLC LLM now supports the newly released model Qwen2! Run it effortlessly on a $100 OrangePi. With Qwen2 0.5B 17.5 tok/s, 1.5B 8.9 tok/s, AI capabilities are more accessible than ever. Explore more at MLC LLM https://t.co/FRLb1NxM7X #MLC #LLM #Qwen2 #OrangePi https://t.co/6aI9bZkGAc https://t.co/fOlaFJZmrY
Meet Qwen2-72B: An Advanced AI Model With 72B Parameters, 128K Token Support, Multilingual Mastery, and SOTA Performance https://t.co/5cxN7cjD9j #AI #Qwen2-72B #multilingual #technology #businesstransformations #ai #news #llm #ml #research #ainews #innovation #artificialintel… https://t.co/DMGe79rtUH
Run Qwen2-72B on clouds & k8s with SkyPilot: 🚀 sky launch serve-72b.yaml See Qwen's official docs for deploying Qwen2 with SkyPilot: https://t.co/SEHrRfQyki Congrats @JustinLin610 & team on the Qwen2 release and becoming #7 open model in Chinese on @lmsysorg Chatbot Arena! https://t.co/zdyBoyBVWn
🚀Alibaba's new Qwen-2 is an open-source model rivaling Metas Llama3! 🤖🌟 I dig the logo ⤵ https://t.co/Qfk1uqkvcq
🆕Qwen2 is available now in low-bit quantized open LLM leaderboard! Check it out - https://t.co/fgeNLUTCee
Qwen-2 with an MMLU of 84.32 is pretty much a GPT-4o /Turbo class model on this benchmark. Here is finetuning code of Qwen-2 with @UnslothAI https://t.co/rm7tKEH6rp
The Qwen research group just dropped an OS 💥This is very very impressive What's new? - Best base model at the LLM Leaderboard + a bunch of benchmarks - 5 base+instruct models in different sizes - The smallest one, of 0.5B params, can even run on-device/locally in your browser… https://t.co/uP9DAj8NwK
Wow! This thing is flying! This is a 2-bit quantized model of Qwen-7B-Instruct! It can’t get any smaller! You can open this directly from @huggingface into @LMStudioAI. 🚀 Congrats to @Alibaba_Qwen and the whole team! 👏🏽💙 https://t.co/SdyW8a6SBW https://t.co/iyuGZWJ1Jo
Uploaded 4bit BnB quants for Qwen2 0.5+1.5+7 & 72b to https://t.co/DY3NvI2F2o @UnslothAI QLoRA finetuning for Qwen2 is 2x faster, uses 70% less VRAM + get 4x longer contexts than FA2! 72b fits in 48GB! Free Colab to finetune 7b: https://t.co/kNsLjAIAXf Ty @JustinLin610 4 Qwen!
🚀 Wow! Very impressive results for Qwen2. 🤯 Nearly 84 MMLU and 85 GSM8K score! Congrats @Alibaba_Qwen for this amazing models! https://t.co/pxTsoMkgKp
🔥Glad to see Qwen2 testing on MixEval and leading the pack on MixEval and other key LLM evaluations. We updated the full results for Qwen2-72B-Instruct on our MixEval leaderboard (shown below). 🤔Impressively, Qwen2-72B-Instruct outperforms Llama-3-70B-Instruct on both… https://t.co/0xf5e7Nl8k https://t.co/RnBxfttk8Z
Awesome to see Qwen2 open-sourced with MLX support out-of-the-box. Pre-quantized MLX models are in the Qwen2 HF repo: https://t.co/Pcz9vQRnvI https://t.co/L8RMBjlRGu https://t.co/I39EVVPV5a
Now it is OFFICIAL! BTW, it's MMLU score is VERY close to gpt4 (86.9) I don't wanna talk too much, but this is the SOTA in open source models. So glad to be working with Eric and @LucasAtkins7 on enabling this. Thanks @Alibaba_Qwen for the excellent base model! https://t.co/mVM6WZAnCt
There you have it! @Alibaba_Qwen just dropped the new Qwen2 base and instruct and their quantized models! Get them on @huggingface https://t.co/kl80agrOEF https://t.co/7ly4cGqcH5
New very, very cool models by Qwen! Results are on the Open LLM Leaderboard, and they combine good scores with - multilinguality (25+ languages!) - code/math capabilities - an Apache 2.0 license for the small models!! Super good job @Alibaba_Qwen ! https://t.co/6kl57Eyqlr
Let's fucking go! Qwen 2 72B 🔥 > Beats Llama 3 70B > Apache 2.0 license (except 72B) > Excels at Code and Math too > 128K context window > AWQ, GPTQ & GGUFs available > 7B beats Llama3 8B and GLM 4 ⚡ Congratulations Qwen team - knocked it outta the park! 🚀 P.S. They are… https://t.co/D4PKndhstz
💗Hello Qwen2! Happy to share the Qwen2 models to you all! 📖 BLOG: https://t.co/0UNwRo1Iea 🤗 HF collection: https://t.co/z6oWkw7Kzb 🤖 https://t.co/Bp56AqQpQJ 💻 GitHub: https://t.co/sEIRe4IDBJ We have base and Instruct models of 5 sizes, Qwen2-0.5B, Qwen2-1.5B, Qwen2-7B,… https://t.co/y5HAu8HcTH
After months of efforts, we are pleased to announce the evolution from Qwen1.5 to Qwen2. This time, we bring to you: ⭐ Base and Instruct models of 5 sizes, including Qwen2-0.5B, Qwen2-1.5B, Qwen2-7B, Qwen2-57B-A14B, and Qwen2-72B. Having been trained on data in 27 additional… https://t.co/SVVxmwwUJ8