---
title: "#quantization articles | Hashnode"
description: "Explore 60 articles tagged #quantization on Hashnode. Tutorials, guides, and how-tos from the community."
url: "https://hashnode.com/tag/quantization"
---

Tag

# #quantization

60 posts · 0 followers

[Abdus Salam](https://hashnode.com/@aheadkit) · [4d ago](https://hashnode.com/posts/running-llms-on-a-2013-ipad-and-a-tv-box-what-smolmaxxing-actually-measured/6af8ffeb7ccdea7e832d2422)

### [Running LLMs on a 2013 iPad and a TV Box: What smolmaxxing Actually Measured](https://aheadkit.hashnode.dev/running-llms-on-a-2013-ipad-and-a-tv-box-what-smolmaxxing-actually-measured?utm_source=hashnode&utm_medium=feed)

A first-generation iPad Air from 2013 has 1 GB of RAM, a two-core Apple A7 chip, and no business running a modern language model. A Xiaomi TV Box S, the kind of thing that sits under a television and

10 min read · [aheadkit.hashnode.dev](https://aheadkit.hashnode.dev/running-llms-on-a-2013-ipad-and-a-tv-box-what-smolmaxxing-actually-measured?utm_source=hashnode&utm_medium=feed)

[Aleksei Romanov](https://hashnode.com/@gfactor) · [Nov 5](https://hashnode.com/posts/quantization-explained-from-fp32-to-int4-and-what-each-precision-costs-you/6aed0294c8a8a0842336cfe1)

### [Quantization, Explained: From FP32 to INT4, and What Each Precision Costs You](https://gfactor.hashnode.dev/quantization-explained-from-fp32-to-int4-and-what-each-precision-costs-you?utm_source=hashnode&utm_medium=feed)

In LoRA & DoRA and our other posts on large models, we keep running into the same wall: memory. This post is about the number formats behind that wall. When you run a modern language model, memory is

14 min read · [gfactor.hashnode.dev](https://gfactor.hashnode.dev/quantization-explained-from-fp32-to-int4-and-what-each-precision-costs-you?utm_source=hashnode&utm_medium=feed)

[Aman Karki](https://hashnode.com/@amankarki) · [Oct 17](https://hashnode.com/posts/blockwise-fp8-ada-kernel-4x-too-slow/6ad2ef38195837f31546215e)

### [Blockwise FP8 on Ada GPUs, part 2: my kernel was correct and 4x too slow](https://amankarki.hashnode.dev/blockwise-fp8-ada-kernel-4x-too-slow?utm_source=hashnode&utm_medium=feed)

How I found register spills in the main loop, fixed them, and got a 1.42x end-to-end speedup on an L4

8 min read · [amankarki.hashnode.dev](https://amankarki.hashnode.dev/blockwise-fp8-ada-kernel-4x-too-slow?utm_source=hashnode&utm_medium=feed)

[Muhammad Tahir](https://hashnode.com/@mtahir27) · [Oct 12](https://hashnode.com/posts/model-distillation-quantization-running-8b-parameter-models-at-a-fraction-of-cloud-cost/6acc7232421a63140113fb51)

### [Model Distillation & Quantization: Running 8B Parameter Models at a Fraction of Cloud Cost](https://mtdeveloper.hashnode.dev/model-distillation-quantization-running-8b-parameter-models-at-a-fraction-of-cloud-cost?utm_source=hashnode&utm_medium=feed)

This article was originally published on Muhammad Tahir's Portfolio. Introduction & Industry Context In the production AI landscape of 2026, enterprise architectures are undergoing a decisive paradigm

13 min read · [mtdeveloper.hashnode.dev](https://mtdeveloper.hashnode.dev/model-distillation-quantization-running-8b-parameter-models-at-a-fraction-of-cloud-cost?utm_source=hashnode&utm_medium=feed)

[Aman Karki](https://hashnode.com/@amankarki) · [Oct 2](https://hashnode.com/posts/blockwise-fp8-ada-gpus-marlin-fallback/6abfad8204a9ca58d66a76ac)

### [Blockwise FP8 on Ada GPUs: why your L4 falls back to Marlin, measured](https://amankarki.hashnode.dev/blockwise-fp8-ada-gpus-marlin-fallback?utm_source=hashnode&utm_medium=feed)

What 128×128 block scaling is, why it needs its own kernel, and what the missing kernel costs on a real L4

8 min read · [amankarki.hashnode.dev](https://amankarki.hashnode.dev/blockwise-fp8-ada-gpus-marlin-fallback?utm_source=hashnode&utm_medium=feed)

[Pranshu Bansal](https://hashnode.com/@pranshubansal) · [Oct 1](https://hashnode.com/posts/i-tested-qdrant-quantization-and-mmap-on-ev-charger-search/6abe241d512cd299a68261c6)

### [I Tested Qdrant Quantization and mmap on EV Charger Search](https://pranshubansal.hashnode.dev/i-tested-qdrant-quantization-and-mmap-on-ev-charger-search?utm_source=hashnode&utm_medium=feed)

What got smaller, what stayed accurate, and where the memory really went

23 min read · [pranshubansal.hashnode.dev](https://pranshubansal.hashnode.dev/i-tested-qdrant-quantization-and-mmap-on-ev-charger-search?utm_source=hashnode&utm_medium=feed)

[market](https://hashnode.com/@eworldlive) · [Sep 28](https://hashnode.com/posts/int8-matched-float32-recall-5-on-our-100-chunk-cart-faq-ubinary-needed-rescore/6aba8e0df409e7fcd0463fbf)

### [Int8 matched float32 Recall@5 on our 100-chunk cart FAQ — ubinary needed rescore](https://eworldlive.hashnode.dev/int8-matched-float32-recall-5-on-our-100-chunk-cart-faq-ubinary-needed-rescore?utm_source=hashnode&utm_medium=feed)

We quantized all-MiniLM-L6-v2 on a fixed 100-chunk ScalarCart FAQ + 15 graded queries (same float32 baseline). No HF MTEB copy-paste. Desk API spend: $0. precision index bytes vs f32 P@5 Recall@5 m

1 min read · [eworldlive.hashnode.dev](https://eworldlive.hashnode.dev/int8-matched-float32-recall-5-on-our-100-chunk-cart-faq-ubinary-needed-rescore?utm_source=hashnode&utm_medium=feed)

[Mihai\_LeanZero](https://hashnode.com/@mihaileanzero) · [Sep 19](https://hashnode.com/posts/mxfp8-vs-q8-10x-the-weight-error-1-the-perplexity/6aae5ab74f3e50fa9189dbd9)

### [MXFP8 vs Q8: 10x the weight error, 1% the perplexity](https://leanzero.hashnode.dev/mxfp8-vs-q8-10x-the-weight-error-1-the-perplexity?utm_source=hashnode&utm_medium=feed)

MXFP8 reconstructs Qwen3-Coder weights about 10x worse than 8-bit affine at identical size, then costs only 1% perplexity. Measured on an M3 Ultra.

20 min read · [leanzero.hashnode.dev](https://leanzero.hashnode.dev/mxfp8-vs-q8-10x-the-weight-error-1-the-perplexity?utm_source=hashnode&utm_medium=feed)

[Rahul Sai Indeevar V](https://hashnode.com/@rahul-ai) · [Sep 16](https://hashnode.com/posts/parameter-efficient-fine-tuning-from-lora-to-qlora/6aaa2c781ddc873f8d269492)

### [Parameter-Efficient Fine-Tuning: From LoRA to QLoRA](https://rahul-ai.hashnode.dev/parameter-efficient-fine-tuning-from-lora-to-qlora?utm_source=hashnode&utm_medium=feed)

Fine-tuning Large Language Models (LLMs) used to be an option only for organizations with massive compute budgets. Full fine-tuning requires updating and storing every parameter, gradient, and optimiz

10 min read · [rahul-ai.hashnode.dev](https://rahul-ai.hashnode.dev/parameter-efficient-fine-tuning-from-lora-to-qlora?utm_source=hashnode&utm_medium=feed)

[mamonu](https://hashnode.com/@mamonu) · [Sep 12](https://hashnode.com/posts/what-fits-and-what-works/6aa4a07d508f0b7ea408dce7)

### [What Fits and What Works](https://mamonu.hashnode.dev/what-fits-and-what-works?utm_source=hashnode&utm_medium=feed)

gpt-oss-20b is my current go-to locally hosted LLM. I explain why

14 min read · [mamonu.hashnode.dev](https://mamonu.hashnode.dev/what-fits-and-what-works?utm_source=hashnode&utm_medium=feed)
