`build_from_module()`¶

Created: 2026-03-18 · Last updated: 2026-03-18

Build an ONNX ModelPackage from a module instance and config.

from mobius import build_from_module

Signature¶

def build_from_module(
    module: nn.Module,
    config: BaseModelConfig,
    task: str | ModelTask = "text-generation",
) -> ModelPackage:

Parameters¶

Parameter	Type	Default	Description
`module`	`nn.Module`	(required)	An `onnxscript.nn.Module` instance. Its `forward()` signature must be compatible with the task.
`config`	`BaseModelConfig`	(required)	Architecture configuration. The `dtype` field controls precision.
`task`	`str \| ModelTask`	`"text-generation"`	Task name string or `ModelTask` instance.

Returns¶

ModelPackage — A dict-like collection of named ir.Model objects.

Examples¶

from mobius import build_from_module, ArchitectureConfig
from mobius.models import CausalLMModel

config = ArchitectureConfig(
    vocab_size=32000,
    max_position_embeddings=4096,
    hidden_size=4096,
    intermediate_size=11008,
    num_hidden_layers=32,
    num_attention_heads=32,
    num_key_value_heads=32,
    hidden_act="silu",
    head_dim=128,
    pad_token_id=0,
)

module = CausalLMModel(config)
pkg = build_from_module(module, config)
pkg["model"]  # ir.Model

Behavior¶

Validates config (if validate() is available)
Casts module parameters to target dtype
Resolves the task and builds the ONNX graph
Applies optimization passes (identity elimination, CSE, etc.)