Xuefei Ning retweeted
Excited to share our recent work GenMAC: a multi-agent collaboration framework for compositional text-to-video generation.
Project page: karine-h.github.io/GenMAC/
Paper: arxiv.org/pdf/2412.04440 huggingface.co/papers/2412.0โฆ
@KaiyiHUANG84276 @yukun6414 @FiNingm @lin_zinan
Xuefei Ning retweeted
[๐๐ป๐๐ฒ๐ฟ๐ป ๐๐ถ๐ฟ๐ถ๐ป๐ด]We are hiring a [๐๐ฉ๐ซ๐ข๐ง๐ ๐๐๐๐] [๐๐ฎ๐ฅ๐ฅ-๐ญ๐ข๐ฆ๐] intern working on ๐ฃ๐ฟ๐ถ๐๐ฎ๐๐ฒ ๐๐๐ผ๐น๐๐๐ถ๐ผ๐ป. If you are interested, please apply here jobs.careers.microsoft.com/gโฆ and send me an email: zinanlin at microsoft dot com
#ICML2024 Spotlight
<Differentially Private Synthetic Data via Foundation Model APIs 2: Text> was accepted at ICML 2024 as Spotlight! Unfortunately, we are not attending ICML in person. But feel free to reach out to us if you are interested!
Paper: arxiv.org/abs/2403.01749
Xuefei Ning retweeted
๐ Image AR models (๐ฉ๐๐ฅ & ๐๐น๐ฎ๐บ๐ฎ๐๐ฒ๐ป) can be distilled to ๐ข๐ก๐ step (up to ๐ฎ๐ญ๐ด๐
๐ณ๐ฎ๐๐๐ฒ๐ฟ) for the first time!
See ๐ซ๐๐๐๐๐๐๐๐
๐ซ๐๐๐๐
๐๐๐ โ
๐ช๐ฒ๐ฏ๐๐ถ๐๐ฒ: imagination-research.github.โฆ
๐ฃ๐ฎ๐ฝ๐ฒ๐ฟ: arxiv.org/abs/2412.17153
huggingface.co/papers/2412.1โฆ
(1/n)
Xuefei Ning retweeted
DiTFastAttn
Attention Compression for Diffusion Transformer Models
Diffusion Transformers (DiT) excel at image and video generation but face computational challenges due to self-attention's quadratic complexity. We propose DiTFastAttn, a novel post-training compression
Xuefei Ning retweeted
Thank @_akhaliq for featuring Skeleton-of-Thoughts! Check out the recorded demo: sites.google.com/view/sot-llโฆ
It is just a start--more work to do towards a usable tool. But we genuinely believe in the potential of this data-driven direction to make LLMs more efficient and powerful!
Skeleton-of-Thought: Large Language Models Can Do Parallel Decoding
paper page: huggingface.co/papers/2307.1โฆ
This work aims at decreasing the end-to-end generation latency of large language models (LLMs). One of the major causes of the high generation latency is the sequential decoding approach adopted by almost all state-of-the-art LLMs. In this work, motivated by the thinking and writing process of humans, we propose "Skeleton-of-Thought" (SoT), which guides LLMs to first generate the skeleton of the answer, and then conducts parallel API calls or batched decoding to complete the contents of each skeleton point in parallel. Not only does SoT provide considerable speed-up (up to 2.39x across 11 different LLMs), but it can also potentially improve the answer quality on several question categories in terms of diversity and relevance. SoT is an initial attempt at data-centric optimization for efficiency, and reveal the potential of pushing LLMs to think more like a human for answer quality.
Xuefei Ning retweeted
Thank @omarsar0 for featuring Skeleton-of-Thoughts! Check out the recorded demo: sites.google.com/view/sot-llโฆ
It is just a start--more work to do towards a usable tool. But we genuinely believe in the potential of this data-driven direction to make LLMs more efficient and powerful!
Skeleton-of-Thought: LLMs can do parallel decoding
Interesting prompting strategy which firsts generate an answer skeleton and then performs parallel API calls to generate the content of each skeleton point.
Reports quality improvements in addition to speed-up of up to 2.39x. Big deal given how costly in terms of latency some tasks are. This a great paper to rethink the necessity of sequential decoding of current LLMs.
arxiv.org/abs/2307.15337
Xuefei Ning retweeted
#ICML Accelerate ๐๐ญ๐๐๐ฅ๐ ๐๐ข๐๐๐ฎ๐ฌ๐ข๐จ๐ง by ๐๐ฑ through a new perspective!
Visit poster #102 on Thursday, July 27th, at 10:30 am.
[๐๐๐ฉ๐๐ซ] OMS-DPM: Optimizing the Model Schedule for Diffusion Probabilistic Models
arxiv.org/abs/2306.08860
(1/3)
Xuefei Ning retweeted
Welcome to my new course on TinyML and Efficient Deep Learning, starting tomorrow: efficientml.ai
Xuefei Ning retweeted
It was awesome giving at tutorial on neural architecture search at #AutoML_Conf with @debadeepta ! See our slides here:
Slides for my latest NAS tutorial with @crwhite_ml at #AutoML_Conf are now online: Part 1: crwhite.ml/assets/nas_tutoriโฆ Part 2: crwhite.ml/assets/nas_tutoriโฆ. Video links coming soon.












