How sujithreddy-47/hybrid-lora-domainadaptation-gpt2 works
This is an academic NLP research project for comparing LoRA, DyLoRA, DC-LoRA, and their hybrid on sequential IMDb-to-Yelp adaptation of GPT-2. The documented workflow runs Domain A training, continues on Domain B, repeats across seeds, and evaluates retention, forgetting, accuracy, latency, and memory. The notebook is the documented execution surface; results and visualizations are saved artifacts. No notebook source was available, so relationships below rely only on README statements and do not claim implementation details beyond them.
Main entry points: code file.ipynb Project notebook · results Result files · images Image outputs
Components 15
Training workflow
- Project notebook
dyloRA-dcLoRA-gpt2-project/code file.ipynb - GPT-2
- IMDb domain
- Yelp domain
- Multi-seed runs
dyloRA-dcLoRA-gpt2-project/code file.ipynb
Adaptation methods
- LoRA
dyloRA-dcLoRA-gpt2-project/code file.ipynb - DyLoRA
dyloRA-dcLoRA-gpt2-project/code file.ipynb - DC-LoRA
dyloRA-dcLoRA-gpt2-project/code file.ipynb - Hybrid method
dyloRA-dcLoRA-gpt2-project/code file.ipynb
Evaluation and analysis
- Evaluation metrics
dyloRA-dcLoRA-gpt2-project/code file.ipynb - Latency and memory
dyloRA-dcLoRA-gpt2-project/code file.ipynb - Visualizations
dyloRA-dcLoRA-gpt2-project/code file.ipynb
Saved outputs
- Result files
dyloRA-dcLoRA-gpt2-project/results - Image outputs
dyloRA-dcLoRA-gpt2-project/images
Other components
- Researcher
Connections 14
- Researcher → Project notebook: runs
README.md - Project notebook → GPT-2: fine-tunes
README.md - Project notebook → IMDb domain: trains on
README.md - Project notebook → Yelp domain: continues on
README.md - Project notebook → LoRA: applies
README.md - Project notebook → DyLoRA: applies
README.md - Project notebook → DC-LoRA: applies
README.md - Project notebook → Hybrid method: applies
README.md - Project notebook → Multi-seed runs: repeats across
README.md - Project notebook → Evaluation metrics: computes
README.md - Project notebook → Latency and memory: profiles
README.md - Project notebook → Visualizations: creates
README.md - Project notebook → Result files: saves
README.md - Visualizations → Image outputs: produces
README.md