HackerTrans
TopNewTrendsCommentsPastAskShowJobs

zhisbug

no profile record

Submissions

[untitled]

1 points·by zhisbug·4 месяца назад·0 comments

[untitled]

1 points·by zhisbug·4 месяца назад·0 comments

Create a 5s 1080p Video in 4.5s with FastVideo on a Single GPU

1080p.fastvideo.org
12 points·by zhisbug·4 месяца назад·1 comments

[untitled]

1 points·by zhisbug·11 месяцев назад·0 comments

[untitled]

1 points·by zhisbug·в прошлом году·0 comments

[untitled]

1 points·by zhisbug·в прошлом году·0 comments

[untitled]

1 points·by zhisbug·в прошлом году·0 comments

[untitled]

1 points·by zhisbug·в прошлом году·0 comments

[untitled]

1 points·by zhisbug·в прошлом году·0 comments

[untitled]

1 points·by zhisbug·в прошлом году·0 comments

Can LLMs play real-time games like supermario (other than Pokemon red)?

twitter.com
3 points·by zhisbug·в прошлом году·1 comments

Sliding Tile Attention: A New Method That Speeds Up HunyuanVideo's Outputs by 3x

old.reddit.com
2 points·by zhisbug·в прошлом году·1 comments

Fast Video Generation with Sliding Tile Attention

hao-ai-lab.github.io
12 points·by zhisbug·в прошлом году·2 comments

More Efficient Chain-of-Thought Reasoning Through Certainty Probing

huggingface.co
6 points·by zhisbug·в прошлом году·2 comments

AI Space Escape: Playing Games While Evaluting LLM Reasonsing

lmgame.org
13 points·by zhisbug·в прошлом году·2 comments

Efficient LLM Scheduling by Learning to Rank

hao-ai-lab.github.io
2 points·by zhisbug·в прошлом году·1 comments

FastVideo: a lightweight framework for accelerating large video diffusion models

github.com
110 points·by zhisbug·2 года назад·24 comments

MuxServe: Flexible Spatial-Temporal Multiplexing for Multiple LLM Serving

hao-ai-lab.github.io
2 points·by zhisbug·2 года назад·1 comments

Consistency LLM: converting LLMs to parallel decoders accelerates inference 3.5x

hao-ai-lab.github.io
461 points·by zhisbug·2 года назад·98 comments

Throughput Is Not All You Need: Maxing Goodput in LLM Serving via Disaggregation

hao-ai-lab.github.io
5 points·by zhisbug·2 года назад·1 comments

comments

zhisbug
·4 месяца назад·discuss
blog: https://haoailab.com/blogs/dreamverse/
zhisbug
·4 месяца назад·discuss
blog: https://haoailab.com/blogs/fastvideo_realtime_1080p/
zhisbug
·11 месяцев назад·discuss
live demo is here: https://fastwan.fastvideo.org/
zhisbug
·в прошлом году·discuss
Pokémon is increasingly used to evaluate modern large language models, but current practices lack standardization, and depend heavily on game-specific harness. The Pokémon Red involves three major tasks—navigation, combat control and training a competitive Pokémon team. We find they come with limitations: navigation tasks are too hard, combat control is too simple, and Pokémon team training is too expensive. We address these issues in Lmgame Bench, a new framework offering standardized evaluations and initial results across diverse games.
zhisbug
·в прошлом году·discuss
where other models tops out in a few moves
zhisbug
·в прошлом году·discuss
More details here: https://x.com/haoailab/status/1909712259326394519
zhisbug
·в прошлом году·discuss
We find that spatial perception and spatial reasoning remain very difficult even for the strongest models like o3 or Claude 3.7
zhisbug
·в прошлом году·discuss
gaming agent code here: https://github.com/lmgame-org/GamingAgent/tree/main
zhisbug
·в прошлом году·discuss
https://hao-ai-lab.github.io/blogs/sta/
zhisbug
·в прошлом году·discuss
Sliding tile attention accelerates Hunyuan video generation by 3x with no quality drop and no need for training
zhisbug
·в прошлом году·discuss
Try our demo and let us know
zhisbug
·в прошлом году·discuss
This is pretty clever and seems to have high potential, but it still relies on humans. What if some day all humans cannot outsmart AI?
zhisbug
·в прошлом году·discuss
please try and give us feedback!
zhisbug
·в прошлом году·discuss
We hope to redefine ai evaluation via our gamified AI evaluation platform: game arena!
zhisbug
·в прошлом году·discuss
We study how to approximate the famous shortest-job-first scheduling in LLM inference!
zhisbug
·2 года назад·discuss
Hugginface model and data link: https://huggingface.co/FastVideo
zhisbug
·2 года назад·discuss
New work from the vLLM team that disaggregates prefill and decoding to maximize goodput (throughput subject to latency constraints) in LLM serving
zhisbug
·3 года назад·discuss
New parallel decoding algorithm that trades flops for latency reduction
zhisbug
·3 года назад·discuss
oops typo
zhisbug
·3 года назад·discuss
yep, it is agonistic to 4-bit. You can deploy a 4-bit model and still use vllm + pagedattention to double or even triple your serving throughput.