neurovlm.core.runtime.NeuroVLMRuntime

class neurovlm.core.runtime.NeuroVLMRuntime(*, spec, model, device, source, checkpoint=None, autoencoder=None, brain_projection=None, text_projection=None, text_encoder=None)

A resolved model pipeline with consistent tensor-level methods.

Use load_pipeline() instead of constructing this class directly. Methods that do not apply to the selected task raise a precise error.

Parameters:

• spec (ModelSpec)

• model (nn.Module)

• device (torch.device)

• source (str)

• checkpoint (Path | None)

• autoencoder (nn.Module | None)

• brain_projection (nn.Module | None)

• text_projection (nn.Module | None)

• text_encoder (TextEncoder | None)

__init__(*, spec, model, device, source, checkpoint=None, autoencoder=None, brain_projection=None, text_projection=None, text_encoder=None)

Parameters:

• spec (ModelSpec)

• model (Module)

• device (device)

• source (str)

• checkpoint (Path | None)

• autoencoder (Module | None)

• brain_projection (Module | None)

• text_projection (Module | None)

• text_encoder (Callable[[Any], Tensor] | None)

Return type:

None

Methods

__init__(*, spec, model, device, source[, ...])
decode(latent)
encode(brain)
encode_brain(brain)
encode_text(text_or_embedding)
generate(text_or_embedding)
reconstruct(brain)
retrieve(brain, text_or_embedding)	Return the full brain-by-text cosine-similarity matrix.
similarity(brain, text_or_embedding)	Return the full brain-by-text cosine-similarity matrix.