--- title: Measuring Model Quantisation Quality with KL Divergence description: KL divergence (KLD) is used to measure how much a quantised LLM deviates from a bf16 reference, providing a better quality signal than perplexity by isolating... canonical: https://daily.dev/posts/measuring-model-quantisation-quality-with-kl-divergence-chfxwppvm twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Measuring Model Quantisation Quality with KL Divergence | daily.dev og:description: KL divergence (KLD) is used to measure how much a quantised LLM deviates from a bf16 reference, providing a better quality signal than perplexity by isolating... og:url: https://daily.dev/posts/measuring-model-quantisation-quality-with-kl-divergence-chfxwppvm og:image: https://api.daily.dev/og/posts/chfxWppVM.png og:image:alt: Measuring Model Quantisation Quality with KL Divergence og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Measuring Model Quantisation Quality with KL Divergence **[Sam McLeod](https://daily.dev/sources/smcleod)** · 13 min read · 1 upvotes · 0 comments ## Summary KL divergence (KLD) is used to measure how much a quantised LLM deviates from a bf16 reference, providing a better quality signal than perplexity by isolating quantisation damage. Using the mlx-kld tool, six MLX quantisations of Qwen 3.6 27B (dense) and five of Qwen 3.6 35B-A3B (MoE) are benchmarked across 65,536 tokens from WikiText-2. Key findings: 8-bit is effectively a bf16 drop-in; 6-bit RTN and oQ6 are statistically tied for the dense model; Unsloth Dynamic 4-bit is a poor deal at 8.6 effective bpw with worse quality than 6-bit; for the MoE model, DWQ-4bit beats oQ4 by ~34% because it protects router tensors; quant rankings don't generalise between dense and MoE architectures. The mlx-kld tool caches reference log-probs as a sparse top-K structure (~420x compression) and loads only one model at a time, making multi-quant comparisons practical on Apple Silicon. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [Do Qwen 3.6 27B quantizations break the pelican?](https://daily.dev/posts/do-qwen-3-6-27b-quantizations-break-the-pelican--gnbqr3e7n) · Quesma · 0 upvotes · 0 comments - [Quantization hurts knowledge nonlinearly](https://daily.dev/posts/quantization-hurts-knowledge-nonlinearly-1k1ale5qk) · Quesma · 1 upvotes · 1 comments - [Q4 vs Q6 vs Q8: The Quantization Decision Framework for Local LLMs](https://daily.dev/posts/q4-vs-q6-vs-q8-the-quantization-decision-framework-for-local-llms-a0umzfovm) · SitePoint · 0 upvotes · 0 comments - [5 LLM Quantization Techniques](https://daily.dev/posts/5-llm-quantization-techniques-nfunrpnuc) · Daily Dose of Data Science \| Avi Chawla \| Substack · 3 upvotes · 1 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#qwen](https://daily.dev/tags/qwen) [View this post on daily.dev](https://daily.dev/posts/measuring-model-quantisation-quality-with-kl-divergence-chfxwppvm) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Measuring Model Quantisation Quality with KL Divergence","url":"https://daily.dev/posts/measuring-model-quantisation-quality-with-kl-divergence-chfxwppvm","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/measuring-model-quantisation-quality-with-kl-divergence-chfxwppvm"},"datePublished":"2026-04-28T04:08:28.170Z","dateModified":"2026-04-30T02:17:09.872Z","description":"KL divergence (KLD) is used to measure how much a quantised LLM deviates from a bf16 reference, providing a better quality signal than perplexity by isolating...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8bfbc08d4cad6032e63b821170c37134?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8bfbc08d4cad6032e63b821170c37134?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Sam McLeod","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Sam McLeod","logo":"https://media.daily.dev/image/upload/s---ERPerFg--/f_auto,q_auto/v1774960030/logos/smcleod?_a=BAMAMiWQ0","url":"https://daily.dev/sources/smcleod"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/measuring-model-quantisation-quality-with-kl-divergence-chfxwppvm","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,qwen","timeRequired":"PT13M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Sam McLeod","item":"https://daily.dev/sources/smcleod"},{"@type":"ListItem","position":3,"name":"Measuring Model Quantisation Quality with KL Divergence"}]} ```