whisper-small-int8-xnnpack-executorch / benchmarks /whisper-small-vivo-x300-fp32.yaml
aorabdel's picture
Sync model repo (text/metadata)
3b61e58 verified
Raw
History Blame Contribute Delete
848 Bytes
version: 1.2.0
report_type: speech-asr
profile: Baseline
created_at: '2026-07-13T00:00:00Z'
context:
target_ref: vivo-x300
runtime:
name: executorch
execution_backend: cpu
config:
n_threads: 4
optimisations:
- XNNPACK
- KleidiAI
dataset:
name: LibriSpeech
slice: test-clean
sample_count: 2620
reference_url: https://huggingface.co/datasets/openslr/librispeech_asr
benchmark:
batch_size: 1
audio_length_s: 7.0
mode: offline
sample_rate_hz: 16000
language: en
num_runs: 50
warmup_runs: 10
performance:
end_to_end_latency_ms:
p50: 10927.0
p90: 11742.0
p99: 11742.0
rtfx: 0.6406
peak_memory_mb: 5226.97
model_load_time_ms: 1327.0
accuracy:
normalised_wer: 3.4454
normalisation: whisper_tokenizer.normalize (Whisper-style)
cer: 1.3323