music/bench/sa3/run_r2_sa3.json
{
"_doc": "Stable Audio 3 Medium bench run. Emitted by harness/music_gen/sa3_generate.py.",
"generator": {
"family": "Stable Audio 3",
"checkpoint": "stable-audio-3-medium",
"model_dir": "D:\\models\\stable-audio-3-medium",
"licence": "Stability AI Community License",
"licence_record": "docs/licence_records/stable_audio_3_medium/LICENSE",
"licence_sha256": "d6f6b1a4dce5c852bd6d7d9482d002baf0ccdb71e662250b73be9eec8764ee8d",
"text_encoder": "t5gemma-b-b-ul2",
"text_encoder_licence": "Gemma Terms of Use",
"text_encoder_licence_record": "docs/licence_records/stable_audio_3_medium/LICENSE_GEMMA",
"text_encoder_licence_sha256": "e77acc0d3163bb7534675045c584b4d04b387b529239fc4b3647da0a01ba4745",
"weights_sha256": "48d9c65e290e7bcd5194e0633bfc2424a59ee9683f5c2d58762d997b7d8ce0b5",
"text_encoder_weights_sha256": "9b05ea5a4f211d023832f706fb2c0e83e4fc721b6da35ab69ceb0b55eb7800d3",
"output_ownership": "OWNED -- Community Licence IV(c)(iii); Gemma 3.3. Model OUTPUT is expressly excluded from the definition of Derivative Work.",
"triggered_obligation": "Community Licence Section III: a Commercial Purpose user MUST register with Stability AI at https://stability.ai/community-license. NOT SATISFIED by this lane -- it is an account action and belongs to Josh.",
"revenue_termination": "Licences terminate above USD $1,000,000 annual revenue; an Enterprise Licence is required from that point.",
"conditioning_inputs_used": [
"text"
]
},
"spec": "build/audio/bench/sa3/spec_r2_sa3.json",
"steps": 8,
"cfg_scale": 1.0,
"half": true,
"ran_at": "2026-08-06T17:26:34+00:00",
"results": [
{
"artifact_id": "MPX_CH02_EXPLORE_r2",
"audio": "D:\\audio\\bench\\sa3\\audio_ear\\MPX_CH02_EXPLORE_r2.wav",
"prompt": "An original instrumental cue for a volcanic crater-lake rim at altitude, cold air off the water, a lone depleted child moving through it with nothing. About 88 seconds long, in 8 clearly different sections. Very sparse: typically only two or three things sound at once, with the upper register left empty for long stretches. The full texture must not arrive until about 22 seconds in. Stop the music completely several times. Roughly 6 percent of the running time is true silence or near silence, held long enough to be heard as a gap and not as a dip. Instrumentation is generic orchestral and untuned percussion only: no gamelan, no gong ensemble, no bonang, no named regional instrument of any living tradition. Instrumental only, no vocals, no singing, no spoken word. 144 BPM",
"bpm_route": "caption_text",
"bpm_appended": "144 BPM",
"seed": 7101,
"seed_route": "passed to generate()",
"prompt_hash": "ad0fbfefbf4c6971f0b2c850287e9cb460385e0828738b9bb699a031a19928bb",
"sha256": "403ae94ef3871121f5787d9363d59337e24f6c00eecdf95776cfd3a445eb9de7",
"timing": {
"generate_s": 4.19,
"handler_load_s": 14.95,
"nvidia_smi_used_mb_before_load": 15242,
"nvidia_smi_used_mb_after_load": 24790,
"nvidia_smi_used_mb_at_record_write": 26704
},
"licence_record": "D:\\audio\\bench\\sa3\\records_ear\\MPX_CH02_EXPLORE_r2.licence.json",
"generation_record": "D:\\audio\\bench\\sa3\\records_ear\\MPX_CH02_EXPLORE_r2.generation.json"
},
{
"artifact_id": "MPX_CH02_BATTLE_r2",
"audio": "D:\\audio\\bench\\sa3\\audio_ear\\MPX_CH02_BATTLE_r2.wav",
"prompt": "An original instrumental cue for the stalk and the fight at the crater-lake waterfall -- a hollowed hunter driven by something that empties him, closing in through volcanic jungle. Change materially about 8 times: each section brings in or removes an element so the seam is audible. Fill the arrangement out: keep six or more layers sounding together for most of the running time. The full texture must not arrive until about 12 seconds in. Keep it continuous; do not leave long empty gaps. Instrumentation is generic orchestral and untuned percussion only: no gamelan, no gong ensemble, no named regional instrument of any living tradition. Instrumental only, no vocals, no singing, no spoken word. 103 BPM",
"bpm_route": "caption_text",
"bpm_appended": "103 BPM",
"seed": 7201,
"seed_route": "passed to generate()",
"prompt_hash": "4a2a774d4ea05862c9ebdffbbe76ffdb4640410b50813203894b6087bba90491",
"sha256": "b3bd0c59f4e8b1ec9ff71c934a01112d6bee1b0efcf56bf703cb9463fee8e39c",
"timing": {
"generate_s": 2.26,
"handler_load_s": 14.95,
"nvidia_smi_used_mb_before_load": 15242,
"nvidia_smi_used_mb_after_load": 24790,
"nvidia_smi_used_mb_at_record_write": 26706
},
"licence_record": "D:\\audio\\bench\\sa3\\records_ear\\MPX_CH02_BATTLE_r2.licence.json",
"generation_record": "D:\\audio\\bench\\sa3\\records_ear\\MPX_CH02_BATTLE_r2.generation.json"
},
{
"artifact_id": "MPX_TITLE_MAIN_r2",
"audio": "D:\\audio\\bench\\sa3\\audio_ear\\MPX_TITLE_MAIN_r2.wav",
"prompt": "An original instrumental cue for a summation over the end credits: the arc's material recalled and answered, warm and forward-moving, an ending that earns its brightness. Change materially about 7 times: each section brings in or removes an element so the seam is audible. Fill the arrangement out: keep six or more layers sounding together for most of the running time. The full texture must not arrive until about 4 seconds in. Energy is held back and peaks about 38 percent of the way through, not at the start. A few short near-silent gaps, about 3 percent of the running time. Instrumental only, no vocals, no singing, no spoken word. 99 BPM",
"bpm_route": "caption_text",
"bpm_appended": "99 BPM",
"seed": 7301,
"seed_route": "passed to generate()",
"prompt_hash": "140519812eeb7e234eb16db58e088ee5810291e91760e830dcd456d3d691d912",
"sha256": "f38b5cda8db405a8b3affc9dd8b9509f9e4bad5109ceff233c62e6367454bc4a",
"timing": {
"generate_s": 2.95,
"handler_load_s": 14.95,
"nvidia_smi_used_mb_before_load": 15242,
"nvidia_smi_used_mb_after_load": 24790,
"nvidia_smi_used_mb_at_record_write": 26706
},
"licence_record": "D:\\audio\\bench\\sa3\\records_ear\\MPX_TITLE_MAIN_r2.licence.json",
"generation_record": "D:\\audio\\bench\\sa3\\records_ear\\MPX_TITLE_MAIN_r2.generation.json"
}
]
}