executorch_dart 0.7.0

executorch_dart: ^0.7.0 copied to clipboard

Added #

• Tokenizer turns text into token ids and back, without a model attached. Encoder models — embeddings, classification, retrieval — feed token ids to forward() and have no generation loop to borrow one from, so until now there was no way to tokenize for them. Supports HuggingFace tokenizer.json built on BPE, SentencePiece, TikToken and llama2.c, detected automatically. Thanks @dariyooo (#45).
  • WordPiece/BERT-family tokenizers are not supported: the ExecuTorch reader is BPE-only and implements no BertNormalizer, which rules out BERT, DistilBERT, MiniLM and most sentence-transformers models. Passing one raises an error naming the specific reason rather than a generic parse failure.
  • Not available on Web, which has no dart:ffi.

Changed #

• Upgraded to ExecuTorch 1.4.0.

Fixed #

• Dependency constraints held hooks, native_toolchain_cmake, and code_assets back from their latest releases. Widened, which also recovers the pub.dev points lost for out-of-date dependencies.

Fixed #

• The example shown on pub.dev now leads with how to use the package — loading a model, running it, and disposing it — instead of notes that only apply inside this repository.

Added #

• First release. Pure-Dart ExecuTorch inference over dart:ffi, extracted from executorch_flutter so Dart servers and command-line programs can run models without a Flutter SDK. Owns the native build hook and the prebuilt binary download.
• executorch_dart_shared.dart (the ffi-free half of the public API) and the ExecutorchManagerBase class are now part of the public API surface. Both exist so executorch_flutter's web implementation can build on this package without reaching into its private internals, and are expected to stay stable like the rest of the public API.