Skip to main content
onyx.load_model(
    name: str,
    version_id: str = None,
    mode: Literal["online", "offline"] = "online"
) -> MLP | RNN | Transformer
Downloads a trained model from the Onyx Engine. Uses local cache when available.

​
Parameters

​
name
str
required
The name of the model to load.
​
version_id
str
default:"None"
The specific version ID to load. If None, loads the latest version.
​
mode
Literal['online', 'offline']
default:"online"
Loading mode:
  • "online": Check Engine for latest version (default)
  • "offline": Load only from local cache without network access

​
Returns

​
model
MLP | RNN | Transformer
The loaded model, ready for inference or simulation. The model is set to evaluation mode (model.eval()).

​
Raises

  • Exception: If the model is not found in the Engine (online mode)
  • Exception: If the model is not found in local storage (offline mode)

​
Example

from onyxengine import Onyx

# Initialize the client
onyx = Onyx()

# Load the latest version
model = onyx.load_model('example_model')

# Check the configuration
print(model.config)
print(f"Architecture: {model.config.type}")
print(f"Inputs: {[i.name for i in model.config.inputs]}")
print(f"Outputs: {[o.name for o in model.config.outputs]}")

​
Load Specific Version

# Load a specific version
model = onyx.load_model(
    'example_model',
    version_id='dcfec841-1748-47e2-b6c7-3c821cc69b4a'
)

​
Offline Mode

# Load from local cache only (no network)
model = onyx.load_model('example_model', mode='offline')

​
Using the Loaded Model

​
Run Simulation

import torch
from onyxengine import Onyx

onyx = Onyx()
model = onyx.load_model('example_model')
seq_length = model.config.sequence_length

# Create inputs
velocity = torch.zeros(1, seq_length, 1)
position = torch.zeros(1, seq_length, 1)
control = torch.ones(1, seq_length + 100, 1)

# Simulate
result = model.simulate(
    x0={'velocity': velocity, 'position': position},
    external_inputs={'control_input': control},
    sim_steps=100
)

​
Direct Inference

import torch
from onyxengine import Onyx

onyx = Onyx()
model = onyx.load_model('example_model')

# Single forward pass
x = torch.randn(1, model.config.sequence_length, len(model.config.inputs))
with torch.no_grad():
    output = model(x)

​
Caching Behavior

Models are cached locally in ~/.onyx/models/:
  • First load downloads the model
  • Subsequent loads use the cache if versions match
  • Specifying a version_id ensures you get that exact version