cera_ffi 0.5.4

Dart bindings for the Cera on-device inference engine. Runs GGUF LLMs locally through dart:ffi, with no Flutter dependency.

example/README.md

Examples

Standalone dart:ffi scripts: a CLI, a server, a test. No Flutter involved, which is the point of this package existing apart from cera_ffi_flutter.

Nothing here bundles a native library, so point the loader at one:

# from the repo root
just dart-libs      # cargo build -p cera-ffi --features ffi-buffer

# from cera_ffi/
dart pub get

CERA_FFI_LIB=../target/debug/libcera_ffi.dylib \
  dart run example/cera_chat.dart /path/to/model.gguf "Why is the sky blue?"

Script	Shows
cera_chat.dart	Chat template, tokenize, generate, decode back to text
cera_generate.dart	Minimal synchronous generate, token IDs only
cera_async.dart	generateAsync + generateStreamingAsync (recommended streaming path)
cera_stream.dart	Synchronous generateStreaming, and why you must drain the event loop
cera_progress.dart	BundleRepo.withProgress download progress callbacks, via fromBundleIdAsync (downloads a full bundle; it cannot be aborted)

They print to stdout by design; analysis_options.yaml disables avoid_print for that reason rather than excluding the directory from analysis, so these still get type-checked.