Skip to content

MapleSilicon/MapleSilicon

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 

Repository files navigation

Maple Silicon Inc.

🌐 Website: https://maplesilicon.co

Maple Silicon Inc. is a Canadian technology company focused on compiler and systems infrastructure for efficient AI computation.

Our work targets a growing gap in modern AI systems: while models increasingly exhibit structured sparsity, today’s software stacks largely treat computation as dense — wasting memory bandwidth, energy, and silicon capacity.

Maple Silicon’s goal is to make sparsity a first-class, verifiable property across the ML stack, starting at the compiler layer.


What We’re Building

SparseFlow (Primary Project)

SparseFlow is an MLIR-based compiler framework designed to preserve, validate, and execute structured sparsity (N:M) across lowering and runtime execution.

Instead of relying on backend-specific kernel tricks or opaque optimizations, SparseFlow treats sparsity as a compiler-level contract:

  • Explicit sparsity metadata
  • Verified transformations
  • Safe fallback to dense execution when constraints are not met

This approach improves correctness, debuggability, and long-term portability across hardware backends.


Why This Matters

Modern AI hardware is constrained by:

  • Memory bandwidth
  • Power limits
  • Inefficient dense execution of sparse models

At the same time:

  • Structured sparsity (e.g. 2:4) is increasingly common in production models
  • Hardware support exists but is fragile and backend-specific
  • End-to-end verification of sparsity is largely missing

SparseFlow addresses this gap at the compiler level, enabling more reliable and transparent sparse execution on existing hardware.


Project Status

  • ✅ CPU correctness validated
  • ✅ Initial GPU functional validation completed
  • ⚠️ GPU kernels not yet performance-tuned
  • 🚧 Benchmarking and cross-vendor validation in progress

Note: SparseFlow is currently in research and development stage.
Performance optimization is not yet the primary goal.


Open Collaboration

Maple Silicon Inc. is open to:

  • Research collaboration
  • Early technical feedback
  • Pilot evaluations
  • Academic and industry discussions

This is foundational infrastructure under active development, not a polished commercial product.


Contact

📧 Email: info@maplesilicon.co
🌐 Website: https://maplesilicon.co


Disclaimer

SparseFlow is an experimental research project.
APIs, behavior, and performance characteristics may change without notice.

About

Compiler and systems infrastructure for efficient AI computation and structured sparsity.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors