Sebastian Raschka corrects torch.manual_seed typo in reasoning model book · Digg

/TECH

Sebastian Raschka corrects torch.manual_seed typo in reasoning model book

Entities: Sebastian Raschka, Build a Reasoning Model (From Scratch)

The change allows readers to reproduce LLaMA-3-8B log-probabilities.

Combined views

39.3K

2 posts, first seen 3h ago

819 likes

Sebastian Raschka corrects torch.manual_seed typo in reasoning model book

Entities: Sebastian Raschka, Build a Reasoning Model (From Scratch)

The change allows readers to reproduce LLaMA-3-8B log-probabilities.

Reactions from ranked influencers

2 posts

SR Sebastian Raschka@rasbt

6:11 AM · Jul 21, 2026

181TECH

Thanks everyone for all the kind words and feedback. Super happy that you are enjoying Build a Reasoning Model (From Scratch)! Unfortunately, there's small typo in listing 6.5 on page 198 (see video below). The line "torch.manual_seed(0)" should be "torch.manual_seed(5)" This correction is needed to reproduce the generated response in listing 6.5 and the corresponding log-probability outputs later in Chapter 6. If you use 0, the generated response and the results that follow will be different. This will be fixed in the next printing. I am sorry about the oversight, and I hope this note saves you some debugging time.

KD Kosta Derpanis (sabbatical in Zurich)@CSProfKGD

6:25 AM · Jul 21, 2026

929TECH

@rasbt Hope it won’t be fixed like the video 😂 Looking forward to purchasing it!

Combined views

39.3K

2 posts, first seen 3h ago

819 likes comments

38 comments

485 saves

52 reposts

Sentiment

Positive95.5%4.5%Negative

Positive Read

Many users thanked Sebastian Raschka for fixing a seed typo in his reasoning model book because the correction supports reproducibility and they value his clear educational content on building models from scratch.

Based on 22 visible X reactions from 37 accounts; directional sample.

Digg Deeper

Ask a question below.

Published answers will appear here.

Reactions from the X community

JZ Jian Zhong@Jian_Zhong_Dev

9:03 AM · Jul 21, 2026

@rasbt Love your Build From Scratch book! I’m looking forward to your future books. Just wondering—do you have any plans to publish a Build From Scratch book on multimodal language models (e.g. VLMs) or vision-language-action (VLA) models?

RK Rajiv Kumar Yadav@createwithrajiv

7:33 AM · Jul 21, 2026

@rasbt @rasbt good catch. for folks reproducing the numbers, is seed=5 required because some later step assumes that exact init, or would any fixed seed work and the text just picked the one matching the reported run?

FC Fabrizio Clipear@FabrizioClipear

7:17 AM · Jul 21, 2026

@rasbt typos in open-source code are just a tax on copy-paste engineering. the real mistake is thinking a seed value changes the lesson.

BA baba@babakarto

6:11 AM · Jul 21, 2026

@rasbt saved my life with this book. thanks goat

View all

CC Craig Certo@craig_certo

6:25 AM · Jul 21, 2026

@rasbt Just ordered the book!

SR Sebastian Raschka@rasbt

6:11 AM · Jul 21, 2026

181TECH

Thanks everyone for all the kind words and feedback. Super happy that you are enjoying Build a Reasoning Model (From Scratch)! Unfortunately, there's small typo in listing 6.5 on page 198 (see video below). The line "torch.manual_seed(0)" should be "torch.manual_seed(5)" This correction is needed to reproduce the generated response in listing 6.5 and the corresponding log-probability outputs later in Chapter 6. If you use 0, the generated response and the results that follow will be different. This will be fixed in the next printing. I am sorry about the oversight, and I hope this note saves you some debugging time.

KD Kosta Derpanis (sabbatical in Zurich)@CSProfKGD

6:25 AM · Jul 21, 2026

929TECH

@rasbt Hope it won’t be fixed like the video 😂 Looking forward to purchasing it!

38

485 saves

52 reposts

Sentiment

Positive95.5%4.5%Negative

Positive Read

Many users thanked Sebastian Raschka for fixing a seed typo in his reasoning model book because the correction supports reproducibility and they value his clear educational content on building models from scratch.

Based on 22 visible X reactions from 37 accounts; directional sample.

Digg Deeper

Ask a question below.

Published answers will appear here.