Can Xu (@CanXu20) / X

Can Xu

71 posts

Can Xu

@CanXu20

Principal Researcher @TencentHunyuan. Ex Founder @WizardLM_AI Creator of WizardLM Family: WizardLM, WizardCoder and WizardMath.

Seattle

Joined August 2021

Pinned
Can Xu
@CanXu20
Jul 28, 2025
🚀 ArtifactsBench v1.1 is here!🔥🔥🔥 Automated visual/frontend code evaluation benchmark with full transparency! 🎯 94.4% consistency with WebDev Arena 🆕 Major v1.1 updates: Added more models @Alibaba_Qwen @Kimi_Moonshot etc 🔓 100% open-source with complete reproducibility
10K
Can Xu
@CanXu20
Jul 15, 2024
🔥 Excited to share the other key Technology of WizardLM-2! 📙AutoEvol: Automatic Instruction Evolving for Large Language Models 🚀We build a fully automated Evol-Instruct pipeline to create high-quality, highly complex instruction tuning data: -------- 🧵 --------
86K
Can Xu
@CanXu20
Sep 23, 2023
We release the code of Evol-Instruct. This is a universal evolution script that can be applied to most domains without modifications. In recent months, we have been busy validating the effectiveness of the first-generation Evol-Instruct in various benchmarks, and we will invest
WizardLM
@WizardLM_AI
Sep 23, 2023
🔥 Excited to open-source the code of our Evol-Instruct！ 🏆It's one the most important techniques for the success of WizardLM, WizardCoder, and WizardMath. ❤️Thanks authors: @CanXu20 , @victorsungo_ai , @tangmensan. 🏇Let's build the Evol Universe Together! 💥Code Link:
11K
Can Xu
@CanXu20
Jan 18, 2024
🎉We have two papers accepted to #ICLR2024! WizardLM (arxiv.org/abs/2304.12244) and WizardCoder (arxiv.org/abs/2306.08568) Congrats to all outstanding co-authors. We will not stop our journey towards AGI. Please stay tuned for our latest work WaveCoder, aiming to create a
4.6K
Can Xu
@CanXu20
Dec 19, 2023
When the data generated by humans in the natural world is gradually exhausted by LLM training, we believe that data deeply processed and created through AI will be the sole avenue towards more powerful AI. Recently, we have spent a lot of effort to re-examine the various problems
WizardLM
@WizardLM_AI
Dec 19, 2023
🔥 Excited to release the SOTA WizardMath-7B-V1.1 model, the best 7B math LLM powered by RLEIF. 🥇 83.2% pass@1 on GSM8k, surpasses ChatGPT 3.5, Gemini Pro, Mixtral, and Claude Instant. 🥇 33.0% pass@1 on MATH, surpasses Mixtral, and comparable with ChatGPT 3.5, Gemini Pro.
4.9K
Can Xu
@CanXu20
Sep 27, 2024
🎉Excited to see the core techniques of WizardLM2 has been accepted to #EMNLP2024 and #NeurIPS2024! 👉Automatic Instruction Evolving for Large Language Models (arxiv.org/abs/2406.00770) 👉Arena Learning: Build Data Flywheel for LLMs Post-training via Simulated Chatbot Arena
5.2K
Can Xu
@CanXu20
Jul 8, 2025
Introducing ArtifactsBench 🎨 An MLLM-as-Judge that evaluates AI-generated UI by looking at a live render. Our pipeline captures interactions via temporal screenshots & scores visual/interactive fidelity against a per-task checklist. This achieves a stunning 94.4% ranking
4.9K
Can Xu
@CanXu20
Aug 29, 2023
We have devoted a great deal of effort to the creation of the Evol-Instruct method and have documented it thoroughly in public papers, with the hope of collaboratively building a prosperous open-source LLM together with everyone. We kindly request the respect of our hard work.
WizardLM
@WizardLM_AI
Aug 29, 2023
Replying to @phindsearch
Please claim your own inconsistent words @phindsearch : 1: You say "Again, we did not use anything from Wizardcoder and I want to make sure that we are extremely clear about this." - So, what's the WizardCoder-style? Can you tell us the details and why use this name-style? 2. As
2.3K
Can Xu
@CanXu20
Jul 11, 2024
Welcome to use Evol-Instruct to build synthetic data pipelines for training large language models In WizardLM2, we have upgraded Evol-Instruct, and we will reveal this "secret" soon🤠
Alpay Ariyak
@AlpayAriyak
Jul 9, 2024
Planning to build a centralized library for synthetic data generation for LLM training(sft first, then pt, rlhf, etc). Will allow you to build pipelines that combine and chain multiple data generation methods together, and modify them individually. Will be housing both
2.1K
Can Xu
@CanXu20
Jul 13, 2024
Our team is dedicated to innovating synthetic training techniques, drawing inspiration from theory of evolution. The previous Evol-Instruct focused on evolving higher-value instructions from the instruction side. This work, Arena Learning, emphasizes the evolution of higher-value
goto
@goto_yuta_
Jul 13, 2024
合成データで学習したLLMをひたすら競争させて、勝ち残ったLLMから新たな合成データを作って学習させて、さらに競争させて...ってループを繰り返すとモデルもデータも高品質になるらしい。この考え方自体に斬新さはあまり感じないけど、「アリーナ学習」っていう厨二感は好み。
4.3K
Can Xu
@CanXu20
Mar 17, 2025
Top 15！ Congratulations to all the collaborators who participated in this interesting competition
Arena.ai
@arena
Mar 17, 2025
New on Arena leaderboard: @Alibaba_Qwen QwQ-32B and @TXhunyuan's HunYuan-TurboS! - Alibaba's QwQ-32B (#12): strong reasoning model with just 32B size. - Tencent's HunYuan-TurboS debuts in the top 15, catching up Zhipu & StepFun. Congrats to both - competition keeps heating up!
3.6K
Can Xu
@CanXu20
Aug 29, 2023
We are introducing models of sizes 7B, 3B, and 1B, catering to users with limited computational power. We will continue to optimize the performance of these smaller-sized models. Please stay tuned to our GitHub github.com/nlpxucan/Wizar… and Hugging Face repositories
WizardLM
@WizardLM_AI
Aug 29, 2023
🎉 Exciting News! Introducing our new member: WizardCoder-Python-7B-V1.0 🚀 🔗 Demo: http://47.103.63.15:50088/ ✨ Offering performance comparable to the renowned WizardCoder-15B-V1.0, this compact version is perfect for those with limited computational resources. Unleash
GitHub - nlpxucan/WizardLM: LLMs build upon Evol Insturct: WizardLM, WizardCoder, WizardMath
From github.com
1K
Can Xu
@CanXu20
Aug 11, 2023
The third member of our wizard family. Foundational science, also holds the key to the path towards AGI. We will continue to make efforts in this field in the future.
WizardLM
@WizardLM_AI
Aug 11, 2023
🔥🔥🔥 Introduce the newest WizardMath models (70B/13B/7B) ! WizardMath 70B achieves: 1. Surpasses ChatGPT-3.5, Claude Instant-1, PaLM-2 and Chinchilla on GSM8k with 81.6 Pass@1 2. Surpasses Text-davinci-002, GAL, PaLM, GPT-3 on MATH with 22.7 Pass@1 3. Surpasses all other
1.3K
Can Xu
@CanXu20
Aug 29, 2023
💥Let's use Evol-Instruct to build the Evol Universe Together
WizardLM
@WizardLM_AI
Aug 28, 2023
💥💥 Wow ! We are happy to know that Phind-CodeLlama-34B is also a "WizardCoder".☺️ As @phindsearch claimed, they used a WizardCoder-style dataset to train their V1 model. And they may continue train WizardCoder to get their V2 model. But there is no need to delete the
8.1K