--- title: Product Key Memory Sparse Coders | daily.dev description: EleutherAI researchers compare Product Key Memory (PKM) transcoders with standard TopK skip transcoders for sparse coding of neural network features. PKM... canonical: https://daily.dev/posts/product-key-memory-sparse-coders-mgaoqj4zr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Product Key Memory Sparse Coders | daily.dev og:description: EleutherAI researchers compare Product Key Memory (PKM) transcoders with standard TopK skip transcoders for sparse coding of neural network features. PKM... og:url: https://daily.dev/posts/product-key-memory-sparse-coders-mgaoqj4zr og:image: https://api.daily.dev/og/posts/MgaOqJ4ZR.png og:image:alt: Product Key Memory Sparse Coders 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. # Product Key Memory Sparse Coders **[EleutherAI](https://daily.dev/sources/eleutherai)** · 7 min read · 0 upvotes · 0 comments ## Summary EleutherAI researchers compare Product Key Memory (PKM) transcoders with standard TopK skip transcoders for sparse coding of neural network features. PKM transcoders split the encoder into two smaller sub-encoders whose combinations map to a larger decoder, reducing encoder parameter count and speeding up training for moderate expansion factors. Experiments on SmolLM2 135M show PKMs train faster than baselines at equivalent latent counts up to a certain size, but larger PKMs (512x expansion) become slower and achieve worse reconstruction. PKMs also score slightly higher on auto-interpretability metrics and produce naturally grouped latents with higher intra-group cosine similarity. However, results are inconsistent across layers, leaving the practical advantage of PKMs over standard sparse transcoders uncertain. Code and models are publicly available. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [5 Embedding Compression Techniques](https://daily.dev/posts/5-embedding-compression-techniques-xrwx2u8js) · Daily Dose of Data Science \| Avi Chawla \| Substack · 1 upvotes · 0 comments - [The Machine Learning Practitioner’s Guide to Speculative Decoding](https://daily.dev/posts/the-machine-learning-practitioner-s-guide-to-speculative-decoding-3z4bou5te) · Machine Learning Mastery · 2 upvotes · 0 comments - [Compressing Streaming Neural Audio Encoders via Latent-Space Distillation](https://daily.dev/posts/compressing-streaming-neural-audio-encoders-via-latent-space-distillation-8tlsmx8nu) · Apple Machine Learning Research · 0 upvotes · 0 comments - [How LLMs Decide What to Forget: KV Cache Eviction Explained](https://daily.dev/posts/how-llms-decide-what-to-forget-kv-cache-eviction-explained-k6kfg62tc) · Medium · 2 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm) [View this post on daily.dev](https://daily.dev/posts/product-key-memory-sparse-coders-mgaoqj4zr) ```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":"Product Key Memory Sparse Coders","url":"https://daily.dev/posts/product-key-memory-sparse-coders-mgaoqj4zr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/product-key-memory-sparse-coders-mgaoqj4zr"},"datePublished":"2026-08-02T07:04:30.353Z","dateModified":"2026-08-02T07:31:39.610Z","description":"EleutherAI researchers compare Product Key Memory (PKM) transcoders with standard TopK skip transcoders for sparse coding of neural network features. PKM...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/7fa3fbd6e99da7e05ab42115e7e68fde?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/7fa3fbd6e99da7e05ab42115e7e68fde?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"EleutherAI","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":"EleutherAI","logo":"https://media.daily.dev/image/upload/s--aAcRyWGF--/f_auto,q_auto/v1785654263/logos/eleutherai","url":"https://daily.dev/sources/eleutherai"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/product-key-memory-sparse-coders-mgaoqj4zr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"EleutherAI","item":"https://daily.dev/sources/eleutherai"},{"@type":"ListItem","position":3,"name":"Product Key Memory Sparse Coders"}]} ```