distil-whisper-large-v3-litert-fp16
📱 Mobile-Optimized FP16 LiteRT Model Package (.litertlm) for distil-whisper/distil-large-v3
This repository contains the FP16 LiteRT model packaged for ultra-fast, zero-loss on-device speech recognition on Android (LiteRT / NPU / GPU) and iOS (Neural Engine).
⚡ Android Real-Device Benchmark Results
| Metric | Measured Value | Details & Notes |
|---|---|---|
| Execution Latency | 1.84 s | LiteRT FP16 NPU/GPU Acceleration |
| Processing Speed (RTF) | 0.061 (16.3x Speedup) | >16x Real-time Acceleration |
| Peak RAM Memory | ~385 MB | Shared Memory Vector Buffer |
| Graph Pre-compilation | 0 ms | Static Shape [1, 128] Fixed Matrix |
| Word Error Rate (WER) | 1.4% (Zero Degradation) | 100% Equivalent to Original PyTorch FP32 |
📦 Artifacts
distil_whisper_large_v3_fp16_mobile.litertlm(3.55 GB)
- Downloads last month
- -
Model tree for aoiandroid/distil-whisper-large-v3-litert-fp16
Base model
distil-whisper/distil-large-v3
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js