🔄 In a Training Loop

Pinkstack

Pinkstack

·

AI & ML interests

LLMs, VLMS, fine-tuning, mid-training, pre-training, reasoning-LLMs, Small(ish) language models. Stacking pink blocks

Recent Activity

published a dataset about 22 hours ago

Pinkstack/RobloxAgent-toolreason-v2-sample100

updated a dataset about 22 hours ago

Pinkstack/RobloxAgent-toolreason-v2-sample100

published a dataset 2 days ago

Pinkstack/RobloxAgent-toolreason-v1-2500

View all activity

Organizations

upvoted a collection 9 months ago

DistilGPT-OSS-qwen3-4B

Qwen3 4b trained on GPT-OSS outputs • 2 items • Updated Dec 31, 2025 • 2

upvoted 9 collections over 1 year ago

SuperThoughts

Trained to reason properly. • 4 items • Updated Mar 2 • 2

Tiny-Parm v2

Small but mighty model with incredible performance for it's small size. • 2 items • Updated Sep 22, 2025 • 3

Parm V1

First gen of the Pinkstack advanced reasoning models. Medium quality but better than the original models they are based on. • 3 items • Updated Sep 22, 2025 • 2

Experimental models

Models which we made as an experiment. • 4 items • Updated Mar 2 • 3

PGAM

Pinkstack general accuracy model(s). Created to hopefully be as accurate as possible. • 1 item • Updated Mar 2 • 3

Parm v2

Advanced chat reasoning models trained on many different datasets, such as QwQ and Sonnet 3.5, these models excel in math, coding, explaining things. • 5 items • Updated Sep 22, 2025 • 4

PRAM v2

Very advanced reasoning models trained on many different datasets, such as qwq and sonnet 3.5, these models excel in math, RP, explaining things. • 4 items • Updated Dec 17, 2024 • 2

PGAM

Pinkstack general accuracy model(s): trained to be accurate using different datasets such as Roblox luau dataset, etc. Not tested for reasoning. • 1 item • Updated Mar 2 • 1

PRAM v1

Pinkstack reasoning accuracy model(s) are generally trained/fine tuned to be better at reasoning than original. V1 version, v2 is better overall. • 3 items • Updated Dec 17, 2024 • 1