whisper-small-ru-v2 GGUF (FP16)

GGUF quantization of artyomboyko/whisper-small-ru-v2 — a fine-tuned Whisper small model for Russian speech recognition.

Format

• Quantization: FP16 (no quality loss vs original F32)
• Size: 483 MB (50% of original 967 MB)
• Architecture: whisper-small (12 encoder layers, 12 decoder layers, 768 hidden dim)
• Sample rate: 16000 Hz (mono)

Usage with whisper.cpp

git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp
wget https://huggingface.co/asmanovlev/whisper-small-ru-v2-gguf/resolve/main/whisper-small-ru-fp16.gguf
make
./main -m whisper-small-ru-fp16.gguf -f audio.wav -l ru

Performance

On Xiaomi 11T (Dimensity 1200, ARMv8.2+ with NEON FP16):

• Real-time transcription with whisper.cpp
• ~2x faster than F32 on Cortex-A78 cores

Credits

• Original model: artyomboyko/whisper-small-ru-v2
• GGUF conversion: whisper.cpp