oxicuda-lm vs llmfit-core vs cortiq-core

Large language model inference primitives for OxiCUDA: BPE tokenizer, transformer layers with KV cache, GPT-2 and LLaMA architectures — pure Rust, zero CUDA SDK dependency.

Core library for llmfit — hardware detection, model fitting, and provider integration

CMF (Cortiq Model Format): a self-describing, memory-mappable binary container for quantized LLM weights, tokenizer, per-task masks and per-skill delta records.