Sub-1-Bit LLM Compression via Latent Factorization(github.com)
70 points by brainless | 12 comments
GitHub repository from SamsungLabs with the official implementation of LittleBit and LittleBit-2, sub-1-bit LLM compression via latent factorization.
#model-compression on Hacker News
5 stories tagged model-compression, newest first.
Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint(prismml.com)
551 points by JonSchneider | 176 comments
PrismML blog post announcing Bonsai 2 27B, a compressed language model claiming near-lossless quality at a 9x smaller footprint.
Bonsai 27B: A 27B-Class model that runs on a phone(prismml.com)
649 points by xenova | 230 comments
PrismML blog post announcing Bonsai 27B, a 27B-parameter-class language model designed to run locally on a phone.
Gemma 4 QAT models: Optimizing compression for mobile and laptop efficiency(blog.google)
384 points by theanonymousone | 120 comments
Google blog post announcing Gemma 4 models trained with Quantization-Aware Training to cut memory use and improve on-device performance on mobile and laptops.
TurboQuant: Redefining AI efficiency with extreme compression(research.google)
495 points by ray__ | 134 comments
Google Research blog post introducing TurboQuant, a method for extreme compression to improve AI model efficiency.