mirage-project/mirage

[Feature Request] - Optimizing Shared Memory Usage in multitoken_paged_attention

Open

#450 opened on Aug 11, 2025

View on GitHub
 (1 comment) (0 reactions) (0 assignees)Cuda (232 forks)auto 404
good first issue

Repository metrics

Stars
 (2,381 stars)
PR merge metrics
 (PR metrics pending)

Description

Feature Description

Our multitoken_paged_attention task cannot process large number of tokens. We should reduce shared memory usage in this task, which would allow us to run prefill using larger batch sizes.

Contributor guide

Research direction
Investigate the multitoken paged attention kernel implementation and identify shared memory usage patterns. Consider techniques like shared memory tiling, register caching, or using larger data types to reduce shared memory pressure. Look at the current kernel launch parameters and try to reduce the shared memory allocation per block while maintaining functionality.
Tech stack
cpp
Domain
backendmachine learningperformance
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
CUDAGPU programmingC++
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40