We finetuned Meta-Llama/Llama-2-7b-hf on the Open-Platypus dataset (garage-bAInd/Open-Platypus) for 5 epochs using MonsterAPI no-code LLM finetuner.
About OpenPlatypus Dataset
OpenPlatypus is focused on improving LLM logical reasoning skills and was used to train the Platypus2 models. The dataset is comprised of various sub-datasets, including PRM800K, ScienceQA, SciBench, ReClor, TheoremQA, among others. These were filtered using keyword search and Sentence Transformers to remove questions with a similarity above 80%. The dataset includes contributions under various licenses like MIT, Creative Commons, and Apache 2.0.
The finetuning session got completed in 1 hour and 30 minutes and costed us only $15 for the entire finetuning run!
Hyperparameters & Run details:
- Model Path: meta-llama/Llama-2-7b-hf
- Dataset: garage-bAInd/Open-Platypus
- Learning rate: 0.0002
- Number of epochs: 5
- Data split: Training: 90% / Validation: 10%
- Gradient accumulation steps: 1
