Downloads · 30 days
6
26% of all-time downloads
whisper-medium-ko
ALLUCY-Rodent/whisper-medium-ko
whisper-medium-ko is a automatic speech recognition model from ALLUCY-Rodent. Use it when you need speech turned into text. The card lists the license as apache-2.0.
seastar105/whisper-medium-komixv2를 optimum 1.27 + onnxruntime 1.29로 익스포트하고 int8(q8) 동적 양자화한 모델 — meeting-ai-pwa의 전사(STT) 기본 모델 승격용.
All-time downloads
23
Public
Repo size
1.2 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.onnx1.2 GB · 100%
At a glance
- Task
- Automatic Speech Recognition
- License
- apache-2.0
- Model type
- whisper
- Access
- Public
- Created
- Aug 26, 2026
- Updated
- Aug 26, 2026
- SHA
- b514bbd2
Base models
Model card
From the Hugging Face model README
whisper-medium-komixv2 (int8 q8, ONNX)
seastar105/whisper-medium-komixv2를 optimum 1.27 + onnxruntime 1.29로 익스포트하고
int8(q8) 동적 양자화한 모델 — meeting-ai-pwa의 전사(STT) 기본 모델 승격용.
파일
| 파일 | 크기 | 비고 |
|---|---|---|
onnx/encoder_model_quantized.onnx | ~314MB | encoder (q8) |
onnx/decoder_model_merged_quantized.onnx | ~873MB | merged decoder (q8) |
| 그 외 | — | tokenizer/config (optimum 익스포트 원본) |
로드 방법 (transformers.js)
import { pipeline } from '@huggingface/transformers';
const asr = await pipeline('automatic-speech-recognition', 'ALLUCY-Rodent/whisper-medium-ko', {
dtype: { encoder_model: 'q8', decoder_model_merged: 'q8' },
});
변환 재현
# 1) 익스포트 (--task 생략 필수 — merged 디코더 미생성 버그)
optimum-cli export onnx -m seastar105/whisper-medium-komixv2 .convert/whisper-medium-komixv2-onnx
# 2) q8 양자화
python scripts/quantize_whisper_ko.py <encoder.onnx> <encoder_q8.onnx>
python scripts/quantize_whisper_merged_decoder.py <decoder_merged.onnx> <decoder_merged_q8.onnx>
A/B 검증: 2026-08-26, AI Hub 단문 3 + 장문 38.8s — small(+11.2%p 정확도 우위).
- Task
- Automatic Speech Recognition
- Type
- whisper
- License
- apache-2.0
- Languages
- ko
- Created
- Aug 26, 2026
- Updated
- Aug 26, 2026