Yizhi Li (@yizhilll) 's Twitter Profile
Yizhi Li

@yizhilll

PhD Student @Manchester_NLP; Multimodal Art Projection research community (m-a-p.ai);

ID: 715820659021062145

linkhttp://yizhilll.github.io calendar_today01-04-2016 08:38:39

211 Tweet

347 Followers

497 Following

Manchester NLP (@manchester_nlp) 's Twitter Profile Photo

We’re pleased to share that the Manchester NLP Group will be presenting *11 papers* at #EMNLP2024. Feel free to drop by and chat with our students and colleagues during their poster sessions and presentations! 🐝🍻 Computer Science @ The University of Manchester @manchester_nlp

We’re pleased to share that the Manchester NLP Group will be presenting *11 papers* at #EMNLP2024. Feel free to drop by and chat with our students and colleagues during their poster sessions and presentations! 🐝🍻
<a href="/csmcr/">Computer Science @ The University of Manchester</a>  @manchester_nlp
Yizhi Li (@yizhilll) 's Twitter Profile Photo

Hi community, I am running the student member of siggen_acl 🤝 Vote for me to build bridges and boost our research community together - "students, industry & academia". I am here to serve YOU. Every vote matters! #NLG #SIGGEN2024 vote at bitl.to/3IQk

Hi community, I am running the student member of  <a href="/siggen_acl/">siggen_acl</a> 🤝 Vote for me to build bridges and boost our research community together - "students, industry &amp; academia". I am here to serve YOU. Every vote matters! #NLG #SIGGEN2024  vote at bitl.to/3IQk
Wenhu Chen (@wenhuchen) 's Twitter Profile Photo

**SoTA-VLM** Vision Language models have been known to be weak at reasoning. Many open source models are inclined to produce very short phrase answers without any intermediate reasoning. One of the major reasons is the lack of CoT-rich instruction datasets. In MAmmoTH-VL, we

**SoTA-VLM**
Vision Language models have been known to be weak at reasoning. Many open source models are inclined to produce very short phrase answers without any intermediate reasoning.

One of the major reasons is the lack of CoT-rich instruction datasets. In MAmmoTH-VL, we
Ge Zhang (@gezhang86038849) 's Twitter Profile Photo

[2/n] 1. Deduplicate: Fineweb undergoes exact deduplication and minhash deduplication. 2. URL Label: Count all the root URLs of fineweb, and use GPT4 to label the top 1 million root URLs in terms of quantity. 3. Broad Recall: Down-sample and recall the data of each domain from

[2/n]
1. Deduplicate: Fineweb undergoes exact deduplication and minhash deduplication.

2. URL Label: Count all the root URLs of fineweb, and use GPT4 to label the top 1 million root URLs in terms of quantity.

3. Broad Recall: Down-sample and recall the data of each domain from
Qian Liu (@sivil_taram) 's Twitter Profile Photo

🎉 Announcing the first Open Science for Foundation Models (SCI-FM) Workshop at #ICLR2025! Join us in advancing transparency and reproducibility in AI through open foundation models. 🤝 Looking to contribute? Join our Program Committee: bit.ly/4acBBjF 🔍 Learn more at:

🎉 Announcing the first Open Science for Foundation Models (SCI-FM) Workshop at #ICLR2025! Join us in advancing transparency and reproducibility in AI through open foundation models.

🤝 Looking to contribute? Join our Program Committee: bit.ly/4acBBjF

🔍 Learn more at:
Ge Zhang (@gezhang86038849) 's Twitter Profile Photo

[1/n] SuperExcited to announce SuperGPQA!!! We spend more than half a year to finally make it done! SuperGPQA is a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines. It also provides the largest human-LLM

[1/n]

SuperExcited to announce SuperGPQA!!!
We spend more than half a year to finally make it done!
SuperGPQA is a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines.
It also provides the largest human-LLM
Haibin (@eric_haibin_lin) 's Twitter Profile Photo

❗️Open source MOE kernels alert❗️ Introducing COMET, a computation/communication library for MoE models from Bytedance. Battle-tested in our 10k+ GPU clusters, COMET shows promising efficiency gains and significant GPU-hour savings (millions 💰💰💰). Integration of DualPipe &

Yizhi Li (@yizhilll) 's Twitter Profile Photo

Amazing!! I try to run it with a very concise prompt to build a website for "predict your recent fortune by your date of birth" and it finished the job perfectly well.

Amazing!! I try to run it with a very concise prompt to build a website for "predict your recent fortune by your date of birth" and it finished the job perfectly well.
Siwei Wu(吴思为) (@siweiwu7) 's Twitter Profile Photo

[1/n] Delighted to share our new work "COIG-P: A High-Quality and Large-Scale Chinese Preference Dataset for Alignment with Human Values". Paper: arxiv.org/abs/2504.05535 HF Daily Paper: huggingface.co/papers/2504.05… Code: github.com/multimodal-art… Data: huggingface.co/collections/m-…

[1/n] Delighted to share our new work "COIG-P: A High-Quality and Large-Scale Chinese Preference Dataset for Alignment with Human Values".
Paper: arxiv.org/abs/2504.05535
HF Daily Paper: huggingface.co/papers/2504.05…
Code: github.com/multimodal-art…
Data: huggingface.co/collections/m-…
Aran Komatsuzaki (@arankomatsuzaki) 's Twitter Profile Photo

Scaling Laws for Native Multimodal Models - Early-fusion exhibits stronger perf at lower param counts, is more efficient to train, and is easier to deploy, compared w/ late fusion. - Incorporating MoEs allows for models that learn modality-specific weights, significantly

Scaling Laws for Native Multimodal Models

- Early-fusion exhibits stronger perf at lower param counts, is more efficient to train, and is easier to deploy, compared w/ late fusion.
- Incorporating MoEs allows for models that learn modality-specific weights, significantly
Ge Zhang (@gezhang86038849) 's Twitter Profile Photo

[1/n] 🚨 Game On for LLM Reasoning—Meet KORGym! 🎮✨ Ever wondered how to truly assess an LLM’s reasoning ability beyond memorized knowledge? Meet our latest breakthrough: KORGym—a dynamic, multi-turn game platform built to reveal the real reasoning skills of language models!

[1/n]
🚨 Game On for LLM Reasoning—Meet KORGym! 🎮✨

Ever wondered how to truly assess an LLM’s reasoning ability beyond memorized knowledge? 

Meet our latest breakthrough: KORGym—a dynamic, multi-turn game platform built to reveal the real reasoning skills of language models!