Automated Training Interface

About

ATI is a standard endpoint layer for computer-led training and inference parameters to prevent the need to rewrite parameter input classes for every model.

Integrate

Add to an existing Python project

pyproject.toml:

[project.dependencies]
...
"ati @ git+https://github.com/darkshapes/ati"
...

[project.scripts]
<your_project_name>_ati_plug = "your_project_name.module:train" # Or whatever

example_module.py

Enhance existing config classes:

>>> from ati import ATISettings
class ExistingConfigClass(ATISettings):

Or build new ones:

>>> from ati import ATISettings                         # Import the template
>>> ati_set = ATISettings(steps=20,dtype="float16")     # Minimum arguments

>>> import torch
>>> import numpy as np
>>> ati_set_torch = ATISettings(steps=20,dtype=torch.float16)       # Framework agnostic
>>> ati_set_numpy = ATISettings(steps=20,dtype=np.dtype(np.float64))

From an Environment File

.env:

steps=20                    # parameters you want to use
dtype="float16"

CLI

template_ati_plug --steps 20 --dtype "float16"      # Same as above

To add a detectable entrypoint to your project make an ati_plug

<your_project_name>_ati_plug --steps 20 --dtype "float16"      # Fill with your project's name

"Agentic" Harness

Use autonomously via .agents/skills/ati/SKILL.md