Learn SparkLang
Tutorials for orchestrating and gating train jobs: dry-run first, then train→status→expect, classify, playbooks, and the IDE. Built for solo and small teams who want reviewable pipelines in PRs without SDK glue.
Tutorial trail
1. Getting Started
Download, install, verify with
--dry-run, configure
dry-run first; optional gateway key for live ask.
2. Your First Program
Write train_eval.spark: train, status, and expect.
3. AI in 5 Minutes
Train→expect, classify, extract, pipeline — the loop in one lesson.
4. Build a Model
model train, status, expect — interactive wizard.
5. Builder factory
SPARK_BC seed, TRAIN/STEP/STATUS opcodes,
--compile /
--run-bc, init weights,
sha256 proofs. Dry ≠ trained.
AI knowledge hive
Engineer explainers — transformers, training, inference, multimodal, agents, evaluation, decompile limits. Cited sources; original diagrams.
Beyond the 5-minute trail
Function Catalog
Search 100+ ops — model, AI, review, browser, network, CUDA, IDE, crypto, and stdlib recipes.
Playground
Edit samples in the browser with catalog-driven presets and dry-run preview.
Language Reference
Full statement spec — dry-run vs live notes per op.
Reference & tools
- Factory hub — full map: compile, decompile, opcodes, build models, train loop, architecture, tokenizer, serve, eval, make, CI/Pages
- AI model aspects — behaviors, ears/STT, eyes/vision, speaking/TTS, thinking, tools, train, eval, serve (status table)
- Voice — listen/speak companions, gated STT/TTS, PSTN defaults off
- Diagrams — tool map, shadows, LLM assist vs Spark deterministic SoT
-
Builder — SPARK_BC
factory E2E: opcodes
0x26/0x28/0x27,--compile/--run-bc, init weights, sha256 proofs; dry ≠ trained - Compile · Decompile · Build models · Train loop
- Architecture · Serve · Eval · Tokenizer
- AI models — analyze, compare, improve, build; dry-run vs gateway
- Language reference — full statement spec
- Programming guide — CLI, build, verified examples
- Playground — edit samples in the browser
- Downloads — installers by platform