server-console

Python Server Console

The Python Server Console is the local browser interface for edge_dit.server. It starts one selected model, sends job requests, shows progress, and displays image or video results.

For the complete installation, shared-library build, virtual environment, and model-path setup, read the Python guide first.

Start the local stack

From this directory:

npm install
EDGE_DIT_FLUX_MODEL_PATH=/absolute/path/to/FLUX.1-dev \
npm run dev:managed

Open http://127.0.0.1:5173. The first model load can take several minutes. The console is usable once the runtime panel reports that the backend is running.

For a video profile, click Apply preset, edit the prompt, and click Create video job. A successful Wan or MiniMax-H3 result can be exported with Save Video as MP4 (ffmpeg must be on the Python Server host). Wan profiles default to 16 fps and MiniMax-H3 defaults to 24 fps. To run again without reloading the model, change the prompt, seed, or other request fields and click Create video job again. MiniMax-H3 must use at least 22 frames and a frame count satisfying 17k+5; its verified preset uses 22.

The command starts three local services:

Service	Address	What it does
Console	http://127.0.0.1:5173	Browser interface and development proxy
Runtime manager	http://127.0.0.1:8090/runtime/v1	Starts/stops profiles and retains a log tail
Python Server	http://127.0.0.1:8080/ed/v2	Runs generation jobs

Every managed profile defaults to auto_allocate=true: stored tensor precision is preserved, and the runtime decides per component what stays resident on the GPU and what streams from CPU using the live free-VRAM budget. The profile manager owns the backend process, so model switching, restart, and runtime log tailing are available in the console.

Use another built-in model profile by setting its model variable and forwarding the profile name:

EDGE_DIT_WAN_VIDEO_MODEL_PATH=/absolute/path/to/Wan2.1-T2V-1.3B-Diffusers \
npm run dev:managed -- --auto-start-profile wan-t2v

The available profiles and their variables are documented in RUNTIME_CONFIGURATION.md. The minimax-h3 profile uses four standalone component variables; MiniMax-H3 requests must contain at least 22 frames and satisfy 17k+5.

Start a custom-policy backend with the console

Use this mode when you need command-line control over quantization or offload. Start the Python Server yourself instead of using dev:managed. For example, this MiniMax-H3 command preserves an existing Q8 DiT and Q4_K_M LLM, keeps the DiT resident when it fits, and stages the text encoder and VAEs from CPU:

conda activate edge
cd /absolute/path/to/edge-dit.cpp

CUDA_VISIBLE_DEVICES=6 python -m edge_dit.server \
  --host 127.0.0.1 \
  --port 8080 \
  --backend cuda \
  --diffusion-model /models/minimax-h3/diffusion_models/minimax-h3-Q8_0.gguf \
  --llm /models/minimax-h3/text_encoders/qwen3vl-minimax-Q4_K_M.gguf \
  --vae /models/minimax-h3/vae/minimax-h3-video-vae-fp16.safetensors \
  --audio-vae /models/minimax-h3/vae/minimax-h3-audio-vae-fp32.safetensors \
  --type preserve \
  --no-auto-allocate \
  --text-encoder-offload \
  --vae-offload \
  --minimax-h3-stage-lifecycle \
  --vae-tiling

Then start only the UI in a second terminal:

conda activate edge
cd /absolute/path/to/edge-dit.cpp/bindings/python/frontend/server-console
npm run dev

Open http://127.0.0.1:5173. Vite proxies /ed/v2 to the direct Python Server on port 8080. Because the Runtime Manager is not running, managed profile start/stop/switch controls and /runtime/v1 status are unavailable; generation, progress, results, repeat generation, and MP4 export still work normally.

Other useful policies are:

--type q8_0 --auto-allocate --max-vram 20
--auto-fit --max-vram 20
--no-auto-allocate --dit-offload --text-encoder-offload --vae-offload
--no-auto-allocate --offload-to-cpu

--type quantizes eligible safetensors while loading. Use --type preserve for pre-quantized GGUF files unless you explicitly want another conversion. --auto-fit chooses TE/DiT quantization as well as placement. See Python bindings and Server for complete semantics and non-MiniMax examples.

Useful commands

npm run dev
npm run runtime:manager -- --auto-start-profile flux-dev
npm run dev:managed:network
npm run build
npm test
npm run test:e2e

npm run dev starts only the UI; a separately running Python Server is required. npm run runtime:manager starts only the model manager. npm run dev:managed:network binds all services to 0.0.0.0 for a trusted local network.

Stop the managed stack with Ctrl-C. The runtime manager also stops the Python Server it created.

Connection targets

The console uses /ed/v2 by default. This is the Python Server HTTP protocol version. The native C++ server has a different /ed/v1 contract, so do not point this console at a native server unless it implements the Python Server job endpoints.

The Console's Connection panel can probe a server running on another machine. For a remote target, use its reachable base URL and the /ed/v2 prefix. The browser must be allowed to reach that address, and the server must be configured for the desired network binding.

Troubleshooting

• The UI opens but the backend is starting: model weights are still loading. Check http://127.0.0.1:8090/runtime/v1/status.
• The manager exits immediately: ensure EDGE_DIT_PYTHON_BIN points to the virtual environment Python and EDGE_DIT_LIBRARY points to libedgedit.so.
• A profile cannot find its model: set the profile's exact EDGE_DIT_*_MODEL_PATH variable to the directory containing model_index.json.
• Port already in use: stop the previous managed stack or override EDGE_DIT_RUNTIME_MANAGER_PORT and EDGE_DIT_MANAGED_BACKEND_PORT. The frontend port is passed through Vite with npm run dev -- --port <port>.