stt-whisper-tiny.en-GGML

Curated Yar deployment artifacts for whisper-tiny.en (English). Part of the Cytognosis model zoo: every artifact carries exact parameter counts, per-quant resource estimates (replaced by measured values at zoo gates), and full provenance.

Requirements manifest (per artifact)

- artifact: ggml-tiny.en-q5_1.bin
  capability: stt
  format: ggml
  quant: q5_1
  disk_bytes: 32166155
  ram_required_mb:
    load: 62
    runtime_peak_est: 90
  exact_params: 37760256
  cpu_floor: arm64-neon | x86-64-avx2
  npu: optional
  license_class: open
  placement_allowed: [device, local_network, own_server]
  latency_class: batch_polish  # live partials come from the streaming engine
  values: estimated  # zoo measurement gate replaces these
- artifact: ggml-tiny.en-q8_0.bin
  capability: stt
  format: ggml
  quant: q8_0
  disk_bytes: 43550795
  ram_required_mb:
    load: 73
    runtime_peak_est: 104
  exact_params: 37760256
  cpu_floor: arm64-neon | x86-64-avx2
  npu: optional
  license_class: open
  placement_allowed: [device, local_network, own_server]
  latency_class: batch_polish  # live partials come from the streaming engine
  values: estimated  # zoo measurement gate replaces these

Provenance

Source weights: openai/whisper-tiny.en (MIT). GGML conversions mirrored from ggerganov/whisper.cpp (MIT). Quantization: upstream whisper.cpp quantize tool. License class: open. Imported by zoo-pipeline/org_setup.py, 2026-07-25. Runtime: whisper.cpp / sherpa-onnx compatible.