{
  "dev": {
    "num_samples": 160,
    "wer": 63.98,
    "cer": 20.63
  },
  "test": {
    "num_samples": 160,
    "wer": 62.83,
    "cer": 19.94
  },
  "_meta": {
    "num_ref_lines": 320,
    "num_hyp_lines": 320,
    "expected_total": 320,
    "used_pairs": 320
  }
}

Notes

This repo is published from scripts/hf_jobs/omni_asr_finetune_recipe.py with PUBLISH_MODEL=1. It preserves the fairseq2 recipe output so the trained checkpoint can be reused by batch evaluators or wrapped for a dedicated Hugging Face Inference Endpoint.

JosueG/omni-asr-llm7b-adja-audiofix2

Omni_AUDIOFIX2_LLM7B_PUBLISH_20260509_R7_llm_1778374727

Source

Evaluation

Notes

JosueG/
omni-asr-llm7b-adja-audiofix2