---
title: "Discussion on \"Temperature, Top-k, and Top-p: How Decoding Changes What a GPT Model Says\n\" | Hashnode"
description: "Discussion on \"Temperature, Top-k, and Top-p: How Decoding Changes What a GPT Model Says\n\". Experiments with temperature, greedy decoding, top-k, and top-p to understand the trade-offs between predictability, repetition, coherence, and creativity."
url: "https://hashnode.com/posts/temperature-top-k-and-top-p-how-decoding-changes-what-a-gpt-model-says/6a6018abd157dc43a1a76a27"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Temperature, Top-k, and Top-p: How Decoding Changes What a GPT Model Says](https://curious-pm.hashnode.dev/temperature-top-k-and-top-p-how-decoding-changes-what-a-gpt-model-says?utm_source=hashnode&utm_medium=feed)

Experiments with temperature, greedy decoding, top-k, and top-p to understand the trade-offs between predictability, repetition, coherence, and creativity.

Jul 22 · 10 min read · curious-pm.hashnode.dev  
Read the article

-   [#temperature](https://hashnode.com/tag/temperature)
-   [#llm](https://hashnode.com/tag/llm)
-   [#deeplearning](https://hashnode.com/tag/deeplearning)

## Written by

[Aman Jain](https://hashnode.com/@aman454)

@aman454 · San Francisco, USA

  
[View profile](https://hashnode.com/@aman454)

LLM enthusiast

Product manager learning to build AI from the inside out — from LLM architecture to product applications.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.