metalSPARSE: A 2:4 Semi-Structured Sparse GEMV Kernel for Apple Silicon GPUs
cs.CE, cs.AI, cs.LG
Submitted: 2026-10-27
Updated: 2026-10-27
Code: https://github.com/Tekraj15/metalSPARS
Terminology
Sources
- Native LLM and MLLM Inference at Scale on Apple Silicon
- Accelerating Sparse Ternary GEMM for Quantized ML on Apple Silicon
- P$^2$ Law: Scaling Law for Post-Training After Model Pruning
- SparseForge: Efficient Semi-Structured LLM Sparsification via Annealing of Hessian-Guided Soft-Mask
- Effective Interplay between Sparsity and Quantization: From Theory to Practice
- PATCH: Learnable Tile-level Hybrid Sparsity for LLMs
- A Proximal Operator for Inducing 2:4-Sparsity
- Rigel: Reverse-Engineering the Metal 4.1 Tensor Compute Path on the Apple M4 Max GPU
- SpenseGPT: Practical One-shot Pruning Enabling Sparse and Dense GEMMs for LLM Inference
- Accelerating Sparse Deep Neural Networks
- Qwen2.5 Technical Report
- BaseRT: Best-in-Class LLM Inference on Apple Silicon via Native Metal
- SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity
- Open-TQ-Metal: Fused Compressed-Domain Attention for Long-Context LLM Inference on Apple Silicon
- A Free Lunch in LLM Compression: Revisiting Retraining after Pruning
- BaseRT: Advancing Best-in-Class LLM Inference with Apple M5 Neural Accelerators
- Wanda++: Pruning Large Language Models via Regional Gradients
- LLM Inference Unveiled: Survey and Roofline Model Insights
Related papers
- Constrained Sensing and Reliable State Estimation with Shallow Recurrent Decoders on a TRIGA Mark II Reactor
- DREChanNet: Patient-Relative iEEG Model for Channel-Wise Epileptogenic Zone Localization
- A matrix-free, differentiable PyTorch solver for phase-field fracture: Formulation, benchmarks, and inverse analysis
- BizAgentBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business
- Colour me shocked: Exact Molecular Hessians from local MLIPs in O(N) time using sparse differentiation!
- In-context learning as a retraining-free paradigm for constitutive modeling