Whisper Small FR (Batch)

Scoring

Primary metric: wer · Tie-breakers: real_time_factor, energy_wh

Metrics: wer, wer_substitutions, wer_deletions, wer_insertions, cer, cer_substitutions, cer_deletions, cer_insertions, real_time_factor, cpu_pct, ram_mb, energy_wh, temperature_c

Changelog

• v1.0.0 — Initial profile (bulk-generated official set).
• v1.1.0 — Declare beam_size/quantization/threads/vad override-eligible (ADR-0009).
• v1.1.1 — Declare wer/cer substitution/deletion/insertion breakdown metrics, now that the runner computes them.
• v1.1.2 — Declare model.context_reset (per_utterance) explicitly -- confirmed empirically (not assumed) that this adapter's Python API already has no cross-utterance context-bleed risk; documents that fact instead of leaving it implicit.

Full definition

{
  "id": "whisper-small-fr-batch",
  "version": "1.1.2",
  "title": "Whisper Small FR (Batch)",
  "benchmark_type": "batch",
  "language": "fr-FR",
  "runtime": {
    "name": "faster-whisper",
    "min_version": "1.0.0"
  },
  "model": {
    "name": "whisper-small",
    "beam_size": 5,
    "temperature": 0,
    "quantization": "int8",
    "vad": true,
    "context_reset": "per_utterance"
  },
  "configuration": {
    "threads": 4
  },
  "normalization": {
    "lowercase": true,
    "remove_punctuation": true,
    "expand_numbers": true,
    "ruleset_id": "goesb-fr-v1"
  },
  "scoring": {
    "primary_metric": "wer",
    "tie_breakers": [
      "real_time_factor",
      "energy_wh"
    ]
  },
  "metrics": [
    "wer",
    "wer_substitutions",
    "wer_deletions",
    "wer_insertions",
    "cer",
    "cer_substitutions",
    "cer_deletions",
    "cer_insertions",
    "real_time_factor",
    "cpu_pct",
    "ram_mb",
    "energy_wh",
    "temperature_c"
  ],
  "overridable": {
    "beam_size": {
      "allowed": [
        1,
        2,
        4,
        5,
        8
      ]
    },
    "vad": {},
    "quantization": {
      "allowed": [
        "int8",
        "float32"
      ]
    },
    "threads": {
      "range": {
        "min": 1,
        "max": 16
      }
    }
  },
  "changelog": [
    {
      "version": "1.0.0",
      "notes": "Initial profile (bulk-generated official set)."
    },
    {
      "version": "1.1.0",
      "notes": "Declare beam_size/quantization/threads/vad override-eligible (ADR-0009)."
    },
    {
      "version": "1.1.1",
      "notes": "Declare wer/cer substitution/deletion/insertion breakdown metrics, now that the runner computes them."
    },
    {
      "version": "1.1.2",
      "notes": "Declare model.context_reset (per_utterance) explicitly -- confirmed empirically (not assumed) that this adapter's Python API already has no cross-utterance context-bleed risk; documents that fact instead of leaving it implicit."
    }
  ]
}