I Made The Smallest (And Dumbest) LLM



I Made ChatGPT-2 Run on a Potato (63MB AI Model!) – Extreme Quantization Experiment
What happens when you compress a billion-parameter AI model down to 63MB? CHAOS.
In this video, I take you through the insane world of AI model quantization – from expensive Ferrari-priced models to a hilariously broken 63MB AI that thinks 1+1 equals… well, you’ll see.
🚀 What You’ll Learn:

Why “open-source” AI models cost more than your house to run
The brutal reality of FP16 vs FP8 vs 4-bit quantization
How to compress GPT-2 into something smaller than your favorite meme
Training AI on pure Twitch chat wisdom and legendary copypastas
The beautiful disaster of 1-bit quantization (spoiler: it’s broken)

Buy Me A Coffee :
https://buymeacoffee.com/codeically

🛠️ Tech Stack Covered:

Hugging Face Transformers
GPT-2 model architecture
llama.cpp compilation nightmares
Custom dataset creation and tokenization
Parallel processing for training optimization

🎯 Keywords:
#AI #MachineLearning #Quantization #GPT2 #OpenSource #TechExperiment #Programming #DeepLearning #ModelCompression #ArtificialIntelligence #HuggingFace #Python #TechTutorial #AIOptimization #ComputerScience

derfully broken that every conversation feels like a fever dream.
🔔 Don’t Miss Out!
If you enjoyed watching me torture innocent AI models in the name of science, SMASH that subscribe button! Your support means everything and helps me create more gloriously chaotic tech content.
📚 Resources Mentioned:
Custom training scripts (link in pinned comment)
#TechComedy #FailedExperiments #AIGoneWrong #BudgetAI #QuantizationFails

first song at 1:15 :
In the Hall of the Mountain King (HD) – From the Soundtrack to “The Social Network”
I do not own this music or intend to infringe it’s copyright.
with the knowledge that the owners of this material have originally released it free of charge

second song at 3:08 :
Pieces Form the Whole (HD) – From the Soundtrack to “The Social Network”
I do not own this music or intend to infringe it’s copyright.
with the knowledge that the owners of this material have originally released it free of charge

source