nn-inplace-rs
| Crates.io | nn-inplace-rs |
| lib.rs | nn-inplace-rs |
| version | 0.1.0 |
| created_at | 2026-01-08 04:56:53.481951+00 |
| updated_at | 2026-01-08 04:56:53.481951+00 |
| description | Optimized NN library for embedded devices. |
| homepage | |
| repository | |
| max_upload_size | |
| id | 2029557 |
| size | 179,510 |
giangndm (giangndm)
documentation
README
Inplace Operation for Neural Network
This repo implements simple neural network inference which optimized for CPU / embeded. Instead of create new tensor everytime, it uses inplace operation to reduce memory usage.
Optimize
Use nkf_aec as example, run 10000 times for measure.
| Version | Fps |
|---|---|
| First | 2649 |
| Vec Dot SIMD & Vec Mul SIMD | 3193 |
| Vec Sigmoid SIMD | 4490 |
| Vec Add, Sub, Mul SIMD | 4560 |
| GRU Fused | 4698 |