Module pytorch 

PyTorch-shaped embedding generation — currently a deterministic MOCK, not real PyTorch inference.

PyTorchEmbedder models the API a real tch (libtorch FFI) or candle-core (Pure Rust) backed embedder would expose, but PyTorchEmbedder::load_model and the internal forward pass never actually load a model or run a neural network: they only validate that model_path exists and then compute a hash-seeded pseudo-random vector. tch requires a C++ libtorch FFI dependency, which is disallowed by the COOLJAPAN Pure Rust Policy for this crate’s default build; candle-core is Pure Rust but is not currently a workspace dependency. Until one of those is wired in, treat this type as test-only / an API placeholder — do not use it for anything that needs real semantic embeddings.

Structs

DeviceManager

Device manager for PyTorch models

PyTorchConfig

PyTorch model configuration

PyTorchEmbedder

PyTorch-shaped embedding generator — currently a deterministic mock, not real PyTorch inference. See the pytorch module-level docs.

PyTorchModelManager

PyTorch model manager for handling multiple models

PyTorchModelMetadata

PyTorch model metadata

PyTorchTokenizer

PyTorch tokenizer for text preprocessing

Enums

ArchitectureType

Neural network architecture types

CompileMode

PyTorch model compilation modes

PyTorchDevice

PyTorch device configuration