faster-whisper-base-int8

INT8-quantized CTranslate2 export of openai/whisper-base, converted for use with faster-whisper.

• Quantization: INT8 (weights compressed from FP32 → INT8; ~75% size reduction)
• Target hardware: CPU, tested down to 2 cores / 4GB RAM
• Source checkpoint: openai/whisper-base

Usage

from faster_whisper import WhisperModel

model = WhisperModel(
    "devxyasir/faster-whisper-base-int8",
    device="cpu",
    compute_type="int8",
    cpu_threads=2,
    num_workers=1,
)

segments, info = model.transcribe("audio.mp3", beam_size=1, vad_filter=True)
for segment in segments:
    print(segment.text)