HackerTrans
TopNewTrendsCommentsPastAskShowJobs

zhisbug

no profile record

Submissions

[untitled]

1 points·by zhisbug·4 mesi fa·0 comments

[untitled]

1 points·by zhisbug·4 mesi fa·0 comments

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

1080p.fastvideo.org
12 points·by zhisbug·4 mesi fa·1 comments

[untitled]

1 points·by zhisbug·11 mesi fa·0 comments

[untitled]

1 points·by zhisbug·anno scorso·0 comments

[untitled]

1 points·by zhisbug·anno scorso·0 comments

[untitled]

1 points·by zhisbug·anno scorso·0 comments

[untitled]

1 points·by zhisbug·anno scorso·0 comments

[untitled]

1 points·by zhisbug·anno scorso·0 comments

[untitled]

1 points·by zhisbug·anno scorso·0 comments

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

twitter.com
3 points·by zhisbug·anno scorso·1 comments

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

old.reddit.com
2 points·by zhisbug·anno scorso·1 comments

Fast Video Generation with Sliding Tile Attention

hao-ai-lab.github.io
12 points·by zhisbug·anno scorso·2 comments

More Efficient Chain-of-Thought Reasoning Through Certainty Probing

huggingface.co
6 points·by zhisbug·anno scorso·2 comments

AI Space Escape: Playing Games While Evaluting LLM Reasonsing

lmgame.org
13 points·by zhisbug·anno scorso·2 comments

Efficient LLM Scheduling by Learning to Rank

hao-ai-lab.github.io
2 points·by zhisbug·anno scorso·1 comments

FastVideo: a lightweight framework for accelerating large video diffusion models

github.com
110 points·by zhisbug·2 anni fa·24 comments

MuxServe: Flexible Spatial-Temporal Multiplexing for Multiple LLM Serving

hao-ai-lab.github.io
2 points·by zhisbug·2 anni fa·1 comments

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

hao-ai-lab.github.io
461 points·by zhisbug·2 anni fa·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 anni fa·1 comments

comments

zhisbug
·4 mesi fa·discuss
blog: https://haoailab.com/blogs/dreamverse/
zhisbug
·4 mesi fa·discuss
blog: https://haoailab.com/blogs/fastvideo_realtime_1080p/
zhisbug
·11 mesi fa·discuss
live demo is here: https://fastwan.fastvideo.org/
zhisbug
·anno scorso·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
·anno scorso·discuss
where other models tops out in a few moves
zhisbug
·anno scorso·discuss
More details here: https://x.com/haoailab/status/1909712259326394519
zhisbug
·anno scorso·discuss
We find that spatial perception and spatial reasoning remain very difficult even for the strongest models like o3 or Claude 3.7
zhisbug
·anno scorso·discuss
gaming agent code here: https://github.com/lmgame-org/GamingAgent/tree/main
zhisbug
·anno scorso·discuss
https://hao-ai-lab.github.io/blogs/sta/
zhisbug
·anno scorso·discuss
Sliding tile attention accelerates Hunyuan video generation by 3x with no quality drop and no need for training
zhisbug
·anno scorso·discuss
Try our demo and let us know
zhisbug
·anno scorso·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
·anno scorso·discuss
please try and give us feedback!
zhisbug
·anno scorso·discuss
We hope to redefine ai evaluation via our gamified AI evaluation platform: game arena!
zhisbug
·anno scorso·discuss
We study how to approximate the famous shortest-job-first scheduling in LLM inference!
zhisbug
·2 anni fa·discuss
Hugginface model and data link: https://huggingface.co/FastVideo
zhisbug
·2 anni fa·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 anni fa·discuss
New parallel decoding algorithm that trades flops for latency reduction
zhisbug
·3 anni fa·discuss
oops typo
zhisbug
·3 anni fa·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.