Pinned Loading

1. faiss-zig faiss-zig Public

   Pure-Zig ANN — Flat, HNSW, IVFFlat, IVFPQ; L2/cosine/inner-product across all families; SIMD @Vector kernels, multi-threaded batch search. 16.94× memory compression on IVFPQ. 76 tests. No C/C++ dep…

   Zig 1

2. inference inference Public

   Pure-Zig LLM serving — paged attention, BF16 kernels, persistent thread pool, safetensors integration. TinyLlama-1.1B end-to-end. 115 tests.

   Zig 1

3. safetensors-zig safetensors-zig Public

   Pure-Zig safetensors reader. @Vector(32,u8) structural scan, BF16/F32/I8. 241µs parse on 201-tensor TinyLlama fixture. 21 tests.

   Zig 1

4. tokenizers-zig tokenizers-zig Public

   Pure-Zig HF tokenizers — BPE, WordPiece, Unigram, full pipeline, offsets, tokenizer.json compat. ~5.3× faster on WordPiece. 189 tests + 600-iter fuzz.

   Zig 1