Description
TurboMul - High-Performance MatMul Solver for Amadeus uPoW
GitHub: https://github.com/himanshu-sugha/TurboMul
Results: 2043 solutions/sec MatMul on N300s RISC-V | Full PoW miner with valid: True, valid_math: True | Successfully validated against Amadeus testnet API
Key Innovation: Through source code analysis of amadeus-utils/src/blake3.rs, discovered Matrix B uses signed int8 (-128 to 127). Combined with float32 optimization, achieved 35x speedup and 100% validation success.
Technical Stack: Python 3.11, NumPy BLAS, Blake3, Koyeb N300s TensTorrent RISC-V, Docker
Note to Judges:
The submission includes a fully functional PoW miner ('miner.py') that connects to the testnet, solves proof-of-work, and submits solutions. Screenshots confirm 'valid_math=True' API responses.
Please check the README.md 'Core Implementation' section for specific code snippets (with line numbers) demonstrating:
1. Correct signed-int8 Matrix B handling
2. Blake3 XOF matrix derivation
3. API chain submission logic
These implementation details prove our solver correctly handles the specific Amadeus protocol logic on RISC-V hardware.
Validation: 100% Correctness (valid_math=True)
Benchmark: 2043 Solutions/Sec (0.489ms latency)
https://github.com/himanshu-sugha/TurboMul