We finetuned Llama2-7B on Code-Alpaca-Instruct Dataset (sahil2801/CodeAlpaca-20k) for 5 epochs or ~ 25,000 steps using MonsterAPI no-code LLM finetuner.
This dataset is HuggingFaceH4/CodeAlpaca_20K unfiltered, removing 36 instances of blatant alignment.
The finetuning session got completed in 4 hours and costed us only $16 for the entire finetuning run!
Hyperparameters & Run details:
- Model Path: meta-llama/Llama-2-7b
- Dataset: sahil2801/CodeAlpaca-20k
- Learning rate: 0.0003
- Number of epochs: 5
- Data split: Training: 90% / Validation: 10%
- Gradient accumulation steps: 1
Loss metrics:
