NeurIPS2024

Code for NeurIPS2024 paper: 'Towards Transparency: Exploring LLM Trainings Datasets through Visual Topic Modeling and Semantic Frames'

Code

The codes for the figures is in the different colab and the figures are in results

• Bunka Package
• DPO Fine-tuning code
• Mistral Classifier with vllm
• Bunka Maps

Datasets:

• Prompt-collective-dataset
• Chatml_dpo_pairs
• Chatml_dpo_pairs Filtered dataset

LLMs Models:

• Base-model ofr benchmark
• Fine-tuned model 1
• Fine-tuned model 2
• LLM for comparison between frames

Embeddings models:

• mxbai-embed-large-v1
• all-MiniLM-L6-v2
• bge-large-en-v1.5
• multi-qa-mpnet-base-dot-v1
• UAE-Large-V1

Benchmarks:

• Open LLM leaderboard

Download model form HF Hub

git-lfs clone https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B