Show HN: Getting GLM 5.2 running on my slow computer

A developer created "colibri," a C-based engine that runs GLM-5.2, a 744-billion parameter mixture-of-experts model, on consumer machines with ~25GB of RAM by keeping only the dense layers in memory and streaming the routed experts from disk on demand. The implementation includes faithful model reproduction validated against the official transformers library, MLA attention with a 57× compressed KV-cache, and native multi-token-prediction speculative decoding with 39-59% draft acceptance rates. The entire engine is a single ~1,300-line C file with no external dependencies, GPU, or Python runtime required.