pictor
Pure Rust 1-bit LLM inference engine for PrismML Bonsai models — umbrella crate.
Status: Stable (thin re-export facade) · Version: 0.2.2 · Updated: 2026-05-30
Re-exports all Pictor subcrates for convenience. Add this single dependency to get access to the entire Pictor ecosystem:
[]
= "0.2.2"
# Enable optional subsystems:
= { = "0.2.2", = ["full"] }
Subcrates
| Crate | Description |
|---|---|
pictor-core |
GGUF loader, tensor types, configuration |
pictor-kernels |
1-bit compute kernels (dequant, GEMV, GEMM, SIMD) |
pictor-model |
Qwen3 transformer family (1.7B/4B/8B), KV cache, attention |
pictor-runtime |
Inference engine, sampling, OpenAI-compatible server |
pictor-tokenizer |
Pure Rust BPE tokenizer (optional) |
pictor-rag |
Retrieval-augmented generation pipeline (optional) |
pictor-eval |
Model evaluation framework (optional) |
pictor-serve |
Standalone OpenAI-compatible server (optional) |
Feature Flags
| Flag | Description |
|---|---|
server |
HTTP server (axum) |
rag |
RAG pipeline |
native-tokenizer |
Pure Rust BPE tokenizer |
eval |
Evaluation harness |
simd-avx2 |
AVX2+FMA SIMD kernels |
simd-avx512 |
AVX-512 SIMD kernels |
simd-neon |
NEON SIMD kernels |
wasm |
WASM-safe build |
full |
All optional features |
License
Apache-2.0 — derived from oxibonsai (COOLJAPAN OU). See LICENSE and NOTICE.