galamkhoahoc/EraX-WoW-Turbo-V1.1-ONNX

EraX WoW Turbo V1.1 — Transformers.js ONNX Q8

Browser-ready ONNX Q8 export of `erax-ai/EraX-WoW-Turbo-V1.1`, the Transformers checkpoint corresponding to `erax-ai/EraX-WoW-Turbo-V1.1-CT2`.

The repository contains the two graphs used by Transformers.js for Whisper sequence-to-sequence generation:

• —onnx/encoder_model_quantized.onnx
• —onnx/decoder_model_merged_quantized.onnx

It also includes the original tokenizer, generation configuration, feature extractor configuration, and Vietnamese normalizer assets.

Browser usage

import { pipeline } from '@huggingface/transformers';

const transcriber = await pipeline(
  'automatic-speech-recognition',
  'galamkhoahoc/EraX-WoW-Turbo-V1.1-ONNX',
  {
    device: 'wasm',
    dtype: {
      encoder_model: 'q8',
      decoder_model_merged: 'q8',
    },
  },
);

const result = await transcriber(audio16k, {
  language: 'vi',
  task: 'transcribe',
});

Model files are downloaded on first use and can be reused through the browser Cache API. Audio inference runs locally in the user's Web Worker and does not need to be sent to an application backend.

Export details

• —Transformers.js converter: 3.8.1
• —Transformers: 4.49.0
• —PyTorch exporter: 2.6.0
• —ONNX Runtime quantization: Q8
• —Audio input: mono 16 kHz

The generated ONNX graphs passed onnx.checker.check_model(..., full_check=True) before upload.