Skip to main contentCrate modelc
Source - arch
- Shared model-structure inspection for transformer architectures.
- chat_template
- Chat template rendering for transformer models.
- cli
- Command-line parsing and helpers shared with the compiler.
- codegen
- compiler
- config
- Configuration file support for modelc.
- constraint
- Grammar-based constrained decoding.
- containerize
- Generate a minimal Docker/OCI image for a .modelc artifact.
- draft
- Neural draft model for speculative decoding (EAGLE-style).
- generate
- Autoregressive text generation for GPT-2 / LLaMA transformer models.
- json_schema
- Minimal JSON Schema validation for generated output.
- lora
- LoRA (Low-Rank Adaptation) adapter support.
- metal
- Apple Silicon Metal acceleration backend (macOS only).
- model
- onnx_exec
- ONNX graph execution engine.
- pack
- Single-file artifact format for modelc.
- parsers
- prefix_cache
- Prompt-prefix KV cache for autoregressive generation.
- runtime
- Tensor runtime and inference ops used by tests and examples (
matmul, linear, etc.). - serve
- HTTP server for
modelc run — loads a .modelc artifact and serves /info + /infer + /chat + /complete. - store
- Local model package store for
modelc list / modelc pull. - tokenizer
- Byte-level BPE tokenizer for GPT-2 style models.
- CLI_VERSION
- Semver plus short git SHA from
build.rs, shown by modelc --version.