Meowmoji turns a short phrase into an animal emoji. The model runs on the device through LiteRT. The model and LiteRT ship with Meowmoji.
The first match compiles the model for the selected chip, CPU or GPU. match returns a list of emoji and a score, highest first. A higher score is a better fit. Don't forget to close the model when you are done.
Meowmoji is arm64-v8a and needs Android 7 (API 24).
dependencies {
implementation("io.github.za4em:sdk:0.1.0")
}Compose helpers:
dependencies {
implementation("io.github.za4em:sdk-compose:0.1.0")
}Meowmoji(context, Accelerator.GPU).use { meowmoji ->
val matches = meowmoji.match("fluffy cat")
}Choose Accelerator.CPU or Accelerator.GPU. Ops the GPU cannot run stay on the CPU.
val meowmoji = rememberMeowmoji(Accelerator.GPU)
val matches = rememberMatchDebounced(meowmoji, text)rememberMatchLatest scores the current phrase and cancels the previous score. rememberMatchDebounced waits for debounce, then scores. If a score is already running, it finishes.
Run the benchmark on a connected phone, or on Firebase Test Lab with gcloud. AGENTS.md has the steps, including how the results become a report. The report is benchmark/report/index.html. The result lines are in benchmark/runs/.
