signal量子位2026-09-27
Laptop Runs 700B-Parameter GLM! No GPU Needed? SSD as VRAM Goes Viral on GitHub
Colibrì is a pure-C, zero-dependency layered inference framework that has gained 32k stars on GitHub. It runs the 744B-parameter GLM-5.2 on a 25GB RAM laptop without a GPU by storing MoE expert weights on SSD and loading them on demand. The int4-quantized model needs only about 372GB of disk space and works with as little as 16GB RAM, and the framework supports nine model families up to 2.8T-parameter Kimi K3.
- for who
- Developers and researchers who want to run frontier-scale MoE models on consumer hardware
- why now
- Colibrì's new GitHub release lets laptops run 744B models without GPU, a major accessibility shift.
- to do
- Download the prebuilt Colibrì binary, fetch the pre-converted int4 GLM-5.2 model, and run 'coli chat' to interact
key points
- Colibrì is a pure-C framework with 32k GitHub stars and zero dependencies
- Runs 744B GLM-5.2 on 25GB RAM without GPU, using SSD for expert weights
- Supports nine model families up to 2.8T Kimi K3, with LRU caching and prefetch
#large model inference#moe#open source framework
score
score 9 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source