No Saved Kaleidosope: an 100% Jitted Neural Network Coding Language with Pythonic Syntax
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866912037766955008 |
|---|---|
| author | da Rosa, Augusto Seben Angeli, Marlon Daniel Junior, Jorge Aikes Ferreira, Alef Iury Gris, Lucas Rafael Soares, Anderson da Silva Junior, Arnaldo Candido de Oliveira, Frederico Santos Damke, Gabriel Trevisan Sousa, Rafael Teixeira |
| author_facet | da Rosa, Augusto Seben Angeli, Marlon Daniel Junior, Jorge Aikes Ferreira, Alef Iury Gris, Lucas Rafael Soares, Anderson da Silva Junior, Arnaldo Candido de Oliveira, Frederico Santos Damke, Gabriel Trevisan Sousa, Rafael Teixeira |
| contents | We developed a jitted compiler for training Artificial Neural Networks using C++, LLVM and Cuda. It features object-oriented characteristics, strong typing, parallel workers for data pre-processing, pythonic syntax for expressions, PyTorch like model declaration and Automatic Differentiation. We implement the mechanisms of cache and pooling in order to manage VRAM, cuBLAS for high performance matrix multiplication and cuDNN for convolutional layers. Our experiments with Residual Convolutional Neural Networks on ImageNet, we reach similar speed but degraded performance. Also, the GRU network experiments show similar accuracy, but our compiler have degraded speed in that task. However, our compiler demonstrates promising results at the CIFAR-10 benchmark, in which we reach the same performance and about the same speed as PyTorch. We make the code publicly available at: https://github.com/NoSavedDATA/NoSavedKaleidoscope |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2409_11600 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | No Saved Kaleidosope: an 100% Jitted Neural Network Coding Language with Pythonic Syntax da Rosa, Augusto Seben Angeli, Marlon Daniel Junior, Jorge Aikes Ferreira, Alef Iury Gris, Lucas Rafael Soares, Anderson da Silva Junior, Arnaldo Candido de Oliveira, Frederico Santos Damke, Gabriel Trevisan Sousa, Rafael Teixeira Programming Languages Artificial Intelligence Machine Learning 68T07 D.3; I.2; I.4; I.7 We developed a jitted compiler for training Artificial Neural Networks using C++, LLVM and Cuda. It features object-oriented characteristics, strong typing, parallel workers for data pre-processing, pythonic syntax for expressions, PyTorch like model declaration and Automatic Differentiation. We implement the mechanisms of cache and pooling in order to manage VRAM, cuBLAS for high performance matrix multiplication and cuDNN for convolutional layers. Our experiments with Residual Convolutional Neural Networks on ImageNet, we reach similar speed but degraded performance. Also, the GRU network experiments show similar accuracy, but our compiler have degraded speed in that task. However, our compiler demonstrates promising results at the CIFAR-10 benchmark, in which we reach the same performance and about the same speed as PyTorch. We make the code publicly available at: https://github.com/NoSavedDATA/NoSavedKaleidoscope |
| title | No Saved Kaleidosope: an 100% Jitted Neural Network Coding Language with Pythonic Syntax |
| topic | Programming Languages Artificial Intelligence Machine Learning 68T07 D.3; I.2; I.4; I.7 |
| url | https://arxiv.org/abs/2409.11600 |