Help with GPT-2 based Emotion Classifier Training (6 classes) - Python, LLM experience needed
Budget: €2 – €6 EUR
Hello,
I’m working on a school project where I’m building an emotion classifier with 6 classes using English text. The model is based on a pretrained GPT-2, following the tutorial series by Sebastian Raschka (currently chapter 5): https://www.youtube.com/watch?v=4yNswvhPWCQ&list=PLTKMiZHVd_2Licpov-ZK24j6oUnbhiPkm
I’m struggling to train the model properly — the accuracy is stuck around 15%, so something is definitely wrong. I suspect it’s a minor issue, but I’m looking for someone with good Python skills and solid understanding of LLM fine-tuning, ideally familiar with this specific tutorial or similar setups.
The existing code structure and workflow should be preserved. Training can be done locally or on Google Colab. The training process requires GPU support. The tutorial repository from Raschka is used to prepare the environment.
What I need:
- Review and debug the training pipeline
- Advice or fixes to improve accuracy
- Ensure code and workflow structure are maintained
Bonus:
- Experience with GPT-2 fine-tuning and emotion classification
- Familiarity with Sebastian Raschka’s tutorials
Please let me know your experience and approach. Thanks!
I’m working on a school project where I’m building an emotion classifier with 6 classes using English text. The model is based on a pretrained GPT-2, following the tutorial series by Sebastian Raschka (currently chapter 5): https://www.youtube.com/watch?v=4yNswvhPWCQ&list=PLTKMiZHVd_2Licpov-ZK24j6oUnbhiPkm
I’m struggling to train the model properly — the accuracy is stuck around 15%, so something is definitely wrong. I suspect it’s a minor issue, but I’m looking for someone with good Python skills and solid understanding of LLM fine-tuning, ideally familiar with this specific tutorial or similar setups.
The existing code structure and workflow should be preserved. Training can be done locally or on Google Colab. The training process requires GPU support. The tutorial repository from Raschka is used to prepare the environment.
What I need:
- Review and debug the training pipeline
- Advice or fixes to improve accuracy
- Ensure code and workflow structure are maintained
Bonus:
- Experience with GPT-2 fine-tuning and emotion classification
- Familiarity with Sebastian Raschka’s tutorials
Please let me know your experience and approach. Thanks!