$ /Users/king/AI/stable-audio-3/optimized/mlx/.venv/bin/python /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/extra-checks.py
# cwd /Users/king/Desktop/Workspace/SRO/zoen
# exit 0

Generating take 1/1: /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/sfx-tests/nested path/userlocal.wav

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SA3 → MLX  text-to-audio
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
      prompt  'Soft fantasy UI bronze tick and wooden knock, isolated sound effect, no music, no voices.'
         dit  sm-sfx          decoder  same-s    
        σmax  1.00
     seconds  1.5s              steps  8                  seed  1785
   DiT dtype  fp16                cfg  1.0
       T_lat  17 (1.58s → trimmed to 1.5s)

  [1/5] T5Gemma encode ·································   259 ms
        embeds (1, 256, 768) mlx.core.float16
  [2/5] Conditioning ···································     4 ms
        cross_attn (1, 257, 768)   global (1, 768)
  [3/5] DiT — load + sample (8 steps, σmax=1.00)
        load 0.3s
        schedule  1.000 · 0.994 · 0.984 · 0.958 · 0.891 · 0.746 · 0.512 · 0.274 · 0.000
        sample 869 ms  (109 ms/step)
  [4/5] Decoder (same-s, FP32)
        load 102 ms
        decode chunked (chunk=8, ovl=2)  →  33 ms   patches (1, 512, 272)
  [5/5] Unpatch + write WAV ····························     2 ms
        audio (2, 66150)   peak 0.253   rms 0.056

  Peak RAM by stage
    T5Gemma encode  ███████████·············    708 MB
    Conditioning    █████████···············    539 MB
    DiT load        ██████████████··········    878 MB
    DiT sample      ████████████████████████   1.45 GB ←  max
    Decoder load    ███·····················    210 MB
    Decode          ███████·················    438 MB
    Unpatch + WAV   ███·····················    211 MB

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  done   1.64s wall  →  1.5s audio  →  0.92× realtime   peak RAM  1.45 GB   seed 1785
  ▸ saved  /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/sfx-tests/nested path/.sfxgen-jll57e6m/source.wav
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
{"path": "/Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/sfx-tests/nested path/userlocal.wav", "sample_rate": 44100, "channels": 2, "seconds": 0.3, "peak_dbfs": -3.0, "bytes": 79424, "boundary_jump": 0.0}
Generating take 1/1: /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/sfx-tests/nested path/userlocal.wav

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SA3 → MLX  text-to-audio
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
      prompt  'Soft fantasy UI bronze tick and wooden knock, isolated sound effect, no music, no voices.'
         dit  sm-sfx          decoder  same-s    
        σmax  1.00
     seconds  1.5s              steps  8                  seed  1785
   DiT dtype  fp16                cfg  1.0
       T_lat  17 (1.58s → trimmed to 1.5s)

  [1/5] T5Gemma encode ·································   235 ms
        embeds (1, 256, 768) mlx.core.float16
  [2/5] Conditioning ···································     2 ms
        cross_attn (1, 257, 768)   global (1, 768)
  [3/5] DiT — load + sample (8 steps, σmax=1.00)
        load 0.2s
        schedule  1.000 · 0.994 · 0.984 · 0.958 · 0.891 · 0.746 · 0.512 · 0.274 · 0.000
        sample 752 ms  (94 ms/step)
  [4/5] Decoder (same-s, FP32)
        load 74 ms
        decode chunked (chunk=8, ovl=2)  →  33 ms   patches (1, 512, 272)
  [5/5] Unpatch + write WAV ····························     1 ms
        audio (2, 66150)   peak 0.253   rms 0.056

  Peak RAM by stage
    T5Gemma encode  ███████████·············    708 MB
    Conditioning    █████████···············    539 MB
    DiT load        ██████████████··········    878 MB
    DiT sample      ████████████████████████   1.45 GB ←  max
    Decoder load    ███·····················    210 MB
    Decode          ███████·················    438 MB
    Unpatch + WAV   ███·····················    211 MB

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  done   1.35s wall  →  1.5s audio  →  1.11× realtime   peak RAM  1.45 GB   seed 1785
  ▸ saved  /Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/sfx-tests/nested path/.sfxgen-nnuy9gzh/source.wav
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
{"path": "/Users/king/Documents/Codex/2026-10-04/files-pasted-by-the-user-i/work/sfx-tests/nested path/userlocal.wav", "sample_rate": 44100, "channels": 2, "seconds": 0.3, "peak_dbfs": -3.0, "bytes": 79424, "boundary_jump": 0.0}
PASS: user-local command, relative path with spaces, explicit format, environment override, normalization, playback, overwrite and deterministic seed