Configuration
Every orchestrator setting is an environment variable and a flag of the same name: DPPAN_JOB_MAX_QUEUED=128 or --job-max-queued 128 (a flag wins). Node and download settings are environment variables on the machine they apply to. The dashboard lists this deployment's limits under Settings → Limits.
Server
| Variable | Default | |
|---|---|---|
PORT | required | The orchestrator's REST API port |
GATEWAY_PORT | PORT + 1000 | The port nodes dial in on, so they need no open ports of their own |
DPPAN_ADMIN_TOKEN | none | Turns on admin endpoints and the dashboard's Admin tab |
DPPAN_MODELS_CONFIG | config/models.toml (the current directory, else next to the binary) | The model registry |
DPPAN_GGUF_<MODEL> | none | The orchestrator's file for a model, e.g. DPPAN_GGUF_LLAMA3_3_70B for llama3.3:70b: see Sharded download |
DPPAN_MODELS_DIR | ~/.dppan/models | Where pull installs and the orchestrator looks, in bundles/ |
LOG_LEVEL | info | trace, debug, info, warn or error |
HEALTH_INTERVAL_SECS | 10 | How often the orchestrator checks its nodes |
INCLAVATE_NO_BANNER | unset | 1 hides the startup banner, as --no-banner does |
Chat
| Variable | Default | |
|---|---|---|
MAX_TOKENS | 512 | Most tokens a reply may have. A reasoning model spends its thinking from the same budget, so raise it if replies come back empty |
MAX_CONTEXT_TOKENS | 2048 | How much of a conversation is kept; past it, the oldest turns are dropped. 0 keeps all of it. Nodes that set DPPAN_CTX need it at least this high: raise the two together |
DPPAN_RANK_TPS_WEIGHT | 1.0 | Multiplier on throughput (the numerator). Raise it to favour faster nodes more aggressively |
DPPAN_RANK_LATENCY_WEIGHT | 1.0 | Multiplier on latency (the denominator). Set below 1.0 to de-emphasise latency |
Jobs
The queue that speech, music, images, video and understanding run through: Generation jobs.
| Variable | Default | |
|---|---|---|
DPPAN_JOB_MAX_QUEUED | 64 | Waiting jobs allowed; running jobs do not count |
DPPAN_INTERACTIVE_LANE | on | off makes speech, voice cloning and transcription wait for a running music, image or video job on the same machine |
DPPAN_JOB_DEADLINE_MINUTES | 10 | Deadline, from submission, for jobs without a time limit of their own, such as qwen3tts and orpheus speech; 0 for none. Image and video jobs use DPPAN_VISUAL_DEADLINE_MINUTES. A request's deadline_seconds replaces either, up to 24 h |
DPPAN_CLEANUP_INTERVAL_SECS | 300 | How often expired artifacts and jobs are removed; 0 never |
DPPAN_WORKER_ID | kept in DPPAN_ARTIFACT_ROOT | Set it only when that directory does not survive a restart (a container with no volume), or unfinished jobs stay running |
Artifacts
Uploads and results: Artifacts.
| Variable | Default | |
|---|---|---|
DPPAN_ARTIFACT_ROOT | ~/.dppan/artifacts | Where artifacts are stored |
DPPAN_ARTIFACT_MAX_BYTES | 0 (unlimited) | Storage allowed per owner, in bytes |
DPPAN_ARTIFACT_MAX_OBJECTS | 0 (unlimited) | Artifacts allowed per owner |
DPPAN_ARTIFACT_TTL_HOURS | 0 (keep forever) | How long new artifacts are kept |
Speech
More on each: Speech and voice › Configuration.
| Variable | Default | |
|---|---|---|
DPPAN_SPEECH_VOICES | unset | Named clips: name=/path/clip.wav, comma-separated |
DPPAN_SPEECH_TEMPERATURE | 0.9 | Default sampling temperature |
DPPAN_SPEECH_TOP_K | 50 | <= 0 disables it |
DPPAN_SPEECH_TOP_P | 1.0 | >= 1.0 disables it |
DPPAN_SPEECH_REPETITION_PENALTY | 1.05 | 1.0 disables it |
DPPAN_SPEECH_MAX_ATTEMPTS | 1 | Retries for a model that does not stop: see When a model will not stop |
DPPAN_SPEECH_MAX_FRAMES | 512 | Length ceiling per piece (~41 s on qwen3tts, ~44 s on orpheus): see Long text |
DPPAN_SPEECH_SYNC_TIMEOUT_SECS | 120 | How long /v1/audio/speech waits before returning a job ID to poll |
DPPAN_SPEECH_MEMORY_BUDGET_MB | half of RAM | Memory all loaded speech, music, image and video models may hold together |
DPPAN_SPEECH_MODEL | unset | Backbone GGUF. Overrides discovery |
DPPAN_SPEECH_MMPROJ | unset | Its decoder GGUF, required alongside it |
DPPAN_SPEECH_MODEL_ID | qwen3-tts | The id requests use for that pair |
Music
More on each: Music › Configuration.
| Variable | Default | |
|---|---|---|
DPPAN_MUSIC_BUNDLE | unset | Model directories, comma-separated. Unset uses ~/.dppan/models/bundles |
DPPAN_MUSIC_MAX_FRAMES | 250 (10 s) | Longest song, in frames at 25 Hz |
DPPAN_MUSIC_STEPS | 0 (the model's own) | Default solver steps |
DPPAN_MUSIC_THREADS | 0 (the machine's) | CPU threads for music |
DPPAN_MUSIC_SYNC_TIMEOUT_SECS | 900 | How long /v1/audio/music waits before returning a job ID to poll |
Images and video
More on each: Image and video settings › Configuration.
| Variable | Default | |
|---|---|---|
DPPAN_VISUAL_MAX_VRAM_GIB | 0 (what is free) | GPU memory per image or video model, in GiB; a negative value leaves that much free. Unset, a model leaves room for the chat of text nodes on the same machine |
DPPAN_VISUAL_NODES | off | on accepts nodes' image and video models: see Image and video nodes |
DPPAN_VISUAL_PREFER_NODES | on | Jobs that name no place try nodes serving only image and video models before this machine. off tries this machine first |
DPPAN_VISUAL_MAX_FRAMES | 481 (20 s at 24 fps) | Longest video a request may ask for, 1 to 1024. A model trained on shorter clips stops at its own length |
DPPAN_VISUAL_DEADLINE_MINUTES | 240 (4 h) | How long an image job may run, and a video for each part it is made in, 1 to 1440 |
DPPAN_VISUAL_IDLE_MINUTES | 10 | How long an image or video model stays loaded with no job, 1 to 1440. Then its process stops and the next job loads it again |
DPPAN_IMAGES_SYNC_TIMEOUT_SECS | 900 | How long /v1/images/generations and /v1/images/edits wait before returning a job ID to poll |
Nodes
Set on the machine running dppan join. Its flags are in the CLI reference.
| Variable | Default | |
|---|---|---|
ORCHESTRATOR_URL | none | The orchestrator, as --orchestrator |
DPPAN_MODEL | llama3.2:1b | The model, as --model |
GGUF_PATH | none | A model file already downloaded, as --gguf |
GATEWAY_URL | the orchestrator's host, port + 1000 | Where to dial the orchestrator's gateway |
DPPAN_CTX | the model's own | The context each chat session is given memory for. Lower it on a machine short of memory, but not below the orchestrator's MAX_CONTEXT_TOKENS |
DPPAN_VISUAL_MAX_VRAM_GIB | 0 (what is free) | On a node offering image and video models: GPU memory per model, as on the orchestrator |
Downloads
| Variable | Default | |
|---|---|---|
HF_TOKEN | unset | Hugging Face token for gated repositories |