English K_Quantization of LLMs Does Not Disproportionately Diminish Multilingual Performance
This addresses concerns for developers and users of locally deployed LLMs about potential biases in quantization practices affecting non-English language performance, though the findings are incremental as they confirm existing methods do not cause disproportionate harm.
The study investigated whether k_quantization of large language models (LLMs) using English-based importance matrices disproportionately harms multilingual performance, finding no significant degradation in English or Norwegian tasks when testing with matrices in English, Norwegian, and Malayalam on the Llama3.3 70B model evaluated with MixEval.
For consumer usage of locally deployed LLMs, the GGUF format and k\_quantization are invaluable tools for maintaining the performance of the original model while reducing it to sizes deployable with consumer-grade hardware. The number of bits dedicated to each weight from the original model is reduced based on how important they are thought to be during model inference. This importance is arrived at through the application of an 'importance matrix'-a relatively small text document meant to be representative of the LLM's standard use-cases. In the vast majority of quants available online, this document is primarily written in English. It was therefore an open question whether performance on English language tasks was preserved through the sacrifice of multilingual performance and whether it can be preserved with alternate importance matrices. This article investigates these hypotheses by quantizing Llama3.3 70B on importance matrices written in three languages (English, Norwegian, and Malayalam) and evaluating them on the MixEval dataset in both English and Norwegian. All experiments related to yielded non-significant results indicating that current quantization practices do not disproportionately harm multilingual performance.