This is not an official SOLAR Model from Upstage, it's just my attempt at a recreation of these powerful models using the new Llama-3
Further Testing Coming Soon
Where the SOLAR Model used a mix of these datasets:
- c-s-ale/alpaca-gpt4-data (SFT)
- Open-Orca/OpenOrca (SFT)
- in-house generated data utilizing Metamath [2] (SFT, DPO)
- Intel/orca_dpo_pairs (DPO)
- allenai/ultrafeedback_binarized_cleaned (DPO)
I Used:
- llm-wizard/alpaca-gpt4-data
- Crystalcareai/slimorca-dedup-alpaca-100k
- meta-math/MetaMathQA
- Intel/orca_dpo_pairs (ORPO)
- (More DPO Datasets May Be Added)
More Info:
- Developed by: cookinai
- License: apache-2.0
- Finetuned from model : unsloth/llama-3-8b-bnb-4bit
This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.
Special Thanks to Upstage's SOLAR Project for the inspiration behind this model
- Downloads last month
- 7