 zeekay
|
3df9b76309
|
feat: add MLX GPU acceleration for C++ and Python SDKs
Implements Apple Silicon GPU acceleration using MLX framework:
C++ Implementation:
- MLXConsensus class with GPU batch processing
- Adaptive batch processor with automatic size tuning
- Example with benchmarks showing 25-30x speedup
- CMake integration with optional MLX support
Python Implementation:
- MLXConsensusBackend with GPU acceleration
- Neural network model for consensus decisions
- Tested with 359.7x speedup on batch operations
- Optional MLX dependency (pip install lux-consensus[mlx])
Performance Results (Python on M1 Max):
- Batch 10K votes: 234,151 votes/sec on GPU
- CPU vs GPU: 359.7x speedup
- Memory: 17.3 MB peak
Both implementations provide:
- Automatic GPU/CPU fallback
- Memory monitoring
- Batch processing optimization
- Zero-copy operations on Apple Silicon
|
2025-11-10 10:52:30 -08:00 |
|