---
title: "Discussion on \"I Trained an Open Decision Model on a Free GPU in 46 Minutes. On My Tests, It Beat Laya.\" | Hashnode"
description: "Discussion on \"I Trained an Open Decision Model on a Free GPU in 46 Minutes. On My Tests, It Beat Laya.\". Typic: an open 400M decision model that returns a probability per option in one pass. Trained free in 46 min; beat Laya on unseen tasks."
url: "https://hashnode.com/posts/i-trained-an-open-decision-model-on-a-free-gpu-in-46-minutes-on-my-tests-it-beat-laya/6abe69f809b5a81c854b6d98"
---

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

Discussion

# [I Trained an Open Decision Model on a Free GPU in 46 Minutes. On My Tests, It Beat Laya.](https://minarsvn.hashnode.dev/i-trained-an-open-decision-model-on-a-free-gpu-in-46-minutes-on-my-tests-it-beat-laya?utm_source=hashnode&utm_medium=feed)

Typic: an open 400M decision model that returns a probability per option in one pass. Trained free in 46 min; beat Laya on unseen tasks.

Oct 1 · 7 min read · minarsvn.hashnode.dev  
Read the article

-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#nlp](https://hashnode.com/tag/nlp)
-   [#opensource](https://hashnode.com/tag/opensource)
-   [#artificial-intelligence](https://hashnode.com/tag/artificial-intelligence)
-   [#python](https://hashnode.com/tag/python)

## Written by

[Minaruzzaman Shovon](https://hashnode.com/@minarsvn)

@minarsvn

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

## Responses 1

Markdown supported · ⌘↵ to post

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[Oct 9](https://hashnode.com/posts/i-trained-an-open-decision-model-on-a-free-gpu-in-46-minutes-on-my-tests-it-beat-laya/6abe69f809b5a81c854b6d98/comment/6ac88f155e564172d8ba764f)

Putting the request after 7,000 tokens is a useful stress test, especially after training on 384-token inputs. I would vary its position next: beginning, middle, and end of the same filler, while keeping the request and allowed options unchanged.

The current setup always makes the final support request the relevant one. A second test could include a plausible but irrelevant ticket near the end while the real request sits in the middle. That would separate long-input tolerance from a preference for the last task-like passage. Reporting the twenty requests as paired outcomes would make those position effects visible without treating small percentage changes as stable rankings.