DTorch
A direct link to the package : here
DTorch is a package made by a student at Epitech to improve his understanding of pytorch and better his knowledge of ia. It is built on top of numpy which is a scientific framework for math between matrices.
This project run on cpu but still have decent computation time making it usable for model building, optimization and saving/loading.
The tensors created can work with numpy but remember that the gradients will not be calculated if the operation are not in the range of control of the library. Therefore, a usage of the packages methods on the tensors if preferable.
Cuda support may appear in the future and similarly for the dkldnn library that seems to be an excellent when working on CPU.
A direct advantage of using dtorch is it's lightness. The package is currently close to 14 Ko and is fast to load in any project while the use of torch often lead to a slow start.
More about DTorch
The package is organised in the following manner :
| Module | Description |
|---|---|
| dtorch | the base module where other are regrouped |
| functionnal | a gather of all the functions that can be used on jtensors |
| jtensors | a simple module containing the JTensors declaration |
| loss | loss classes for model implementation |
| nn | a neural network layer system and its multiple layers (see doc for more details). |
| einops | a basic reproduction of some einops library methods that support autograd |
| optim | A library contaning the optimizers that can be used on the networks parameters. |
| dtorchvision | A lib containing visual datasets and models (pretrained too) |
| dtorchtext | A lib containing textual datasets and models (pretrained too) |
About Autograd
The implementation in this library may not be the most effective computationnaly but the goal was to make it simple and make it work effectively by relying on numpy matrices operations for computation speed concerns.
1 maintainer
Metadata
1 maintainer
Release files for dtorch 0.0.7
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| dtorch-0.0.7.tar.gz | 23.6 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| dtorch-0.0.7-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 50.5 kB
Release files / dtorch-0.0.7.tar.gz
| Download URL | dtorch-0.0.7.tar.gz |
|---|---|
| Size | 23.6 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
faa77aeb81da6f74e358dfce0ffbe3dd777033426ab86fd33765e83e4dce0ff6
|
|
BLAKE2b-256 checksum How to use checksums |
770041ed95d87a9774b2c6daba7dd130036685c6b2dcf3a72da3ff97f675d91c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.2 CPython/3.11.3
|
Release files / dtorch-0.0.7-py3-none-any.whl
| Download URL | dtorch-0.0.7-py3-none-any.whl |
|---|---|
| Size | 27.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
352be5a7aab1632be4bbaa4cb03ce6424bed20740e0ac6809d6d3d9d92c2a4e1
|
|
BLAKE2b-256 checksum How to use checksums |
486705833c7825d039ba7f7ba1fa30c1acbd012ddae457e172f4ef551d67e283
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/4.0.2 CPython/3.11.3
|
