docs/runtime-startup-fix.json

{
  "date": "2026-09-09",
  "observed_space": "hasimjaneef/OmniAI-GPU",
  "observed_commit": "900c07fbfd925be7d5f15c8390f9697a2ab296e3",
  "observed_stage": "RUNTIME_ERROR",
  "requested_hardware": "a100-large",
  "user_log_startup_utc": "2026-09-09 18:07:11",
  "user_log_checkpoint_shards_loaded": 4,
  "user_log_last_native_message": "audio tokenizer type is set to s3tokenizer_step_audio",
  "underlying_exception": "Unknown: the deployed startup handler suppressed it.",
  "source_revision": "503e754207c94da6bb26850b4469f367c9ea3582",
  "changes": [
    "Retain vision and audio input; disable unused TTS module initialization.",
    "Use an instance-scoped compatibility shim to skip the pinned duplex constructor's unconditional speech-renderer call, restoring the method in finally.",
    "Retain native generate_audio=False and exclude speech-renderer assets from snapshot download.",
    "Report startup stage and exception type/code locations without raw exception messages, URLs, locals or credentials."
  ],
  "checks": {
    "backend_cases_passed": 58,
    "frontend_cases_passed": 36,
    "typescript_and_production_build": "passed",
    "release_validation": "passed",
    "pinned_source_hashes_and_ast": "passed, five source files",
    "exact_pinned_constructor_and_prepare_fixture": "passed with inert decoder/tokenizer/tensor substitutes; no model weights or GPU execution",
    "targeted_regressions_added": 2
  },
  "revised_gpu_startup_verified": false,
  "real_model_inference_verified": false,
  "uploaded_or_restarted_by_codex": false
}