Machine Learning

1 readers

1 users here now

Community Rules:

Be nice. No offensive behavior, insults or attacks: we encourage a diverse community in which members feel safe and have a voice.
Make your post clear and comprehensive: posts that lack insight or effort will be removed. (ex: questions which are easily googled)
Beginner or career related questions go elsewhere. This community is focused in discussion of research and new projects that advance the state-of-the-art.
Limit self-promotion. Comments and posts should be first and foremost about topics of interest to ML observers and practitioners. Limited self-promotion is tolerated, but the sub is not here as merely a source for free advertisement. Such posts will be removed at the discretion of the mods.

founded 1 year ago

MODERATORS

communick@academy.garden

[R] Break the Sequential Dependency of LLM Inference Using Lookahead Decoding (alien.top)

submitted 11 months ago by APaperADay@alien.top to c/machinelearning@academy.garden

1 comments fedilink hide all child comments

Code: https://github.com/hao-ai-lab/LookaheadDecoding

Blog post: https://lmsys.org/blog/2023-11-21-lookahead-decoding/

Description:

We introduce lookahead decoding, a new, exact, and parallel decoding algorithm to accelerate LLM inference. Lookahead decoding breaks the sequential dependency in autoregressive decoding by concurrently extracting and verifying n-grams directly with the LLM, utilizing the Jacobi iteration method. Lookahead decoding functions without the need for a draft model or a data store. It linearly decreases the number of decoding steps directly correlating with the log(FLOPs) used per decoding step. Below is a demo of lookahead decoding accelerating LLaMa-2-Chat 7B generation:

https://i.redd.it/k61qtr4zz22c1.gif

top 1 comments

sorted by: hot top controversial new old

[–] too_long_story@alien.top 1 points 11 months ago

Well, but how to merry it with batching so that flash attention kernels can work with it?

Any complicated masks for attention imply hard times of making possible supporting batching.