Pinned Loading

1. iron-batch iron-batch Public

   LLM serving primitives in Rust: paged KV allocation, continuous batching, streaming HTTP.

   Rust

2. Topolens Topolens Public

   Reads NUMA/PCIe topology from sysfs and recommends CPU-to-accelerator pinning

   Rust

3. bit-floor bit-floor Public

   1-bit and ternary LLM inference engine in Rust and CUDA. Weights stay quantized — no dequantization during inference.

   Rust

4. sift-core sift-core Public

   Hybrid search from scratch in Rust: LSM storage, BM25, flat vector index, RRF fusion. Zero deps.

   Rust