Skip to content

Testing

aivrar edited this page Sep 26, 2026 · 1 revision

Testing

Testing in Omni Studio

Developer controls for selecting installed variants and workers. Loading and inference were not performed for this capture. Captured September 25, 2026.

Testing is a developer tool: load downloaded weights, then send a one-shot prompt and inspect the raw reply. It is not a replacement for Chat sessions, not a music DAW, and not a Comfy queue.

Open the sidebar Developer tools disclosure, then Testing.

Use it when you want to know "does this variant actually answer?" without creating a 100-message session. For multi-turn work, use Chat.

Load an Omni worker

The first card is Load omni model worker.

  1. Model family — chat-capable ids. Moshi, Audio Lab, ACE-Step, MOSS-TTS, and MOSS-SFX are hidden here on purpose. Moshi has no batch path.
  2. Downloaded variant (repo) — only variants you already downloaded. Empty list means go to Model library.
  3. Device
  4. Precision — auto / fp16 / bf16 / fp32
  5. Spawn worker

Banners appear when the family has no weights or no variants. Open Models jumps to Model library.

If a worker already exists for that family, the card shows its status and variant so you do not spawn a duplicate by accident. Kill extras on Runtime.

One-shot inference

Once a worker is ready, the prompt box, temperature, top-p, and max tokens send a single /api/chat/{model} style request (non-session). The raw text lands in the response pane. This is for contract checks (QWEN3B_TEXT_OK style), not for roleplay.

Testing also has Image, Audio, and Video file inputs below the prompt. It sends those files with the one-shot request, subject to the chosen worker's actual modality support. Responses are not streamed. Qwen audio/video and Moshi duplex remain unavailable here.

If the response is an error JSON, read it. 501 on audio/video is the honest Qwen/MiniCPM video behavior, not a broken button.

Audio Lab and ACE-Step loaders

Lower cards let you pick installed Audio Lab SA/CLAP/VAE or ACE-Step DiT/LM/VAE (including custom: names) and load them the same way the audio tabs do. Generate on the Audio Lab or Music tab after loading; Testing has no Audio Lab or ACE-Step generation form.

Prefer the real Audio Lab / Music tabs for anything you want to keep. Those tabs have progress, jobs, LoRAs, and ranked output. Testing is a smoke path.

Still obey:

  • cheap state reads before load
  • explicit device
  • unload + delete the exact worker after the test if residency is not wanted
  • ACE and Audio Lab inference is synchronous; do not poll generic jobs

What Testing will not do

  • It will not create Chat sessions or persist 24h history
  • It will not queue Comfy workflows
  • It will not enable Voice & TTS
  • It will not make Moshi batch-infer (hidden)
  • It will not bypass valid: false on an oversized spawn

Related pages

Clone this wiki locally