Guides & Reference
Everything you need to understand and use ResonTech — from product overview to production integration.
Guides for getting started, running jobs, and integrating your code.
Get started
Inference
Serve vLLM models
How it works
One serve_module.py + one inference.yaml become an OpenAI-compatible endpoint any OpenAI client can call.
serve_module.py
The Python file that declares your model: LLMConfig per engine, exported as app.
inference.yaml
The vLLM-mode yaml — required fields, the vllm: block, and which cluster knobs matter.
LLMConfig options
The knobs that decide whether your model boots and how much context it serves — VRAM math included.
Examples
Ready-to-run vLLM jobs — chat, coding, vision, RAG bundles, ASR — each a folder you submit as-is.
Qwen3-8B guarded RAG chat
Four engines on one 24 GB GPU: chat, moderation guard, embeddings, reranker — behind one endpoint.
Qwen2.5-VL-32B vision
Document vision for hard scans, handwriting, dense tables — structured JSON out.
Hardware requirements
How to read VRAM minimums, what each GPU tier fits, and the CPU / RAM / disk budget per worker.
Serve custom models
How it works
Ship one Python class, get a /predict endpoint — the platform owns HTTP, auth, and scaling.
serve.py
The predictor contract: __init__ + predict(data: bytes) → dict, plus three optional hooks.
inference.yaml
The custom-mode yaml — module, class_name, init_args, weights, and cluster sizing.
Examples
Ready-to-run predictor jobs — segmentation, image generation, TTS, classification.
SAM 2.1 segmentation
Point-prompted image segmentation on any 12 GB card — click a point, get masks.
Training
Submit via UI wizard
Walk through the 7-step wizard to launch a training job from the browser.
Submit via Python SDK
Submit a training job from Python with one rt_submit() call.
Track jobs & download results
Monitor FL rounds, download logs, and grab the final model checkpoint.
FL examples gallery
Real FL recipes spanning vision, language, speech, embeddings, medical, and image generation.
Reference
SDK client config
ResonTechConfig fields — required keys, defaults, env-var pattern, self-hosted overrides.
Training config
TrainingConfig fields, the extra bag, federation knobs, and preview-before-submit.
Submit inference via SDK
rt_submit_inference() in detail — generator mode, BYO bundles, weights, sample data.
Calling /predict
Five client patterns (file / bytes / URL / JSON / batch), the predict_json gotcha, error handling.
Files browser (web UI)
Your auto-provisioned bucket in the browser — upload folders, edit scripts in Monaco.
Storage layout & I/O (SDK)
Bucket layout, upload mechanics, presigned downloads, and re-running an existing workspace.
Custom training classes
Override the model, executor, or persistor — what the contract requires.
Troubleshooting
Common failure modes by what you were trying to do — install, login, submit, predict.
Advanced & theory
How FL works on ResonTech
FedAvg plumbing, the per-round loop, persistor lifecycle, and empirical tuning lessons.
Port a centralized recipe to FL
Already have train.py? Wrap your model + write fl_train_model() to ship it as an FL job.
Dataset format & sharding
manifest.ndjson, zip shard rules, and choosing shard count to match worker count.
Job lifecycle deep-dive
JobState values, polling patterns, output locations, cancellation, and archival.
Infrastructure & clusters
Three GPU cluster types, how the distributed network aggregates compute, hardware floors.
Product overview, company background, and platform policies.