Ge Zhang
@GeZhang86038849
Followers
2K
Following
714
Statuses
638
Founder: M-A-P(https://t.co/CGWz8JrHzH) Contribution @Tiger_Lab @Yi-01.AI @Bytedance MAP-Neo, MMMU, MAmmoTH, MMLU-pro, Yi, COIG, YuE
Joined April 2021
@laurence_ai @xidulu If talking about the brief intuition, broadly speaking, it is indeed MoE. But if you check the mechanism detailedly, it is way more beautiful. And it is practical.
0
0
0
@cognitivecompai check the email affiliation maybe. I have no idea but these talented folks may already have it training in progress.
0
0
0
Thanks for sharing our work!
Generating Symbolic World Models via Test-time Scaling of Large Language Models
0
0
5
Around one week remaining for submission! Share your insights about foundation models with us in SCI-FM @ ICLR 2025!
๐ 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. ๐ค Looking to contribute? Join our Program Committee: ๐ Learn more at: #OpenScience #MachineLearning #FoundationModels 1/N
0
2
10
[7/n] Resources: I personally own very little credit of the paper. Amazing work still, glad to contribute to it and help the Bytedance's engineers. We do notice that Code-LLM has significantly changed how coders work!
0
0
2
@soldni Wish that Yi and deepseek have the support of investigators and are truly with more than 10k H800s. Then MAP may not be soooooo poor as well.
1
0
1
@soldni The rumor is just crazy. At least in the last half of 2024, I donโt think that they own much more GPUs than Yi. These Chinese entrepreneur LLM teams are just poor.
0
0
0
M-A-Pโs Chinese New Year gift to the Open-Source Community. LLaMA moment of music foundation model! The first open-source model with Suno-v3.5 Level performance! Your personal music AI assistant is already here.
1/n: ๐ Announcing YuE (ไน) โ the most powerful open-source full-song music generation model! ๐ต Tackle the lyrics-to-song task (like with support for diverse genres, stunning vocals, & multiple languages. Bonus? Itโs Hugging Face & LLAMA-compatible for easy fine-tuning. ๐ ๏ธ Code: Demo:
0
1
14
@alexandr_wang This manโs success is a representative tragedy of Chinese Americans. African Americans and Jews win their rights by fighting. You folks win by kneeing and faking no bias against you in academics, career, etc while there is too much.
2
0
14
Amazing! Love the work!
๐ Meet EvaByte: The best open-source tokenizer-free language model! Our 6.5B byte LM matches modern tokenizer-based LMs with 5x less data & 2x faster decoding, naturally extending to multimodal tasks while fixing tokenization quirks. ๐ป Blog: ๐งต 1/9
0
0
2
These idiots saying this know nothing about training an LLM and don't read paper. They only know emotional geopolitics instead of rational engineering and science. The cost of Yi-Lightning and MiniMax-01 is no larger and even smaller. Answer is here for a long time.
deepseek is a ccp state psyop + economic warfare to make american ai unprofitable they are faking the cost was low to justify setting price low and hoping everyone switches to it damage AI competitiveness in the us dont take the bait
0
0
9