Quantization matters (via) What impact does quantization have on the performance of an LLM? been wondering about this for quite a while, now here are numbers from Paul Gauthier.
He ran differently quantized versions of Qwen 2.5 32B Instruct through his Aider code editing benchmark and saw a range of scores.
The original released weights (BF16) scored highest at 71.4%, with Ollama's qwen2.5-coder:32b-instruct-fp16 (a 66GB download) achieving the same score.
The quantized Ollama qwen2.5-coder:32b-instruct-q4_K_M (a 20GB download) saw a massive drop in quality, scoring just 53.4% on the same benchmark.
Recent articles
- Weeknotes: asynchronous LLMs, synchronous embeddings, and I kind of started a podcast - 22nd November 2024
- Notes from Bing Chat—Our First Encounter With Manipulative AI - 19th November 2024
- Project: Civic Band - scraping and searching PDF meeting minutes from hundreds of municipalities - 16th November 2024