prompt_strategies.dpo.llama3

prompt_strategies.dpo.llama3

DPO strategies for llama-3 chat template

Functions

Name Description
argilla_chat for argilla/dpo-mix-7k conversations
icr chatml transforms for datasets with system, input, chosen, rejected
intel For Intel Orca DPO Pairs
ultra for ultrafeedback binarized conversations

argilla_chat

prompt_strategies.dpo.llama3.argilla_chat(cfg, **kwargs)

for argilla/dpo-mix-7k conversations

icr

prompt_strategies.dpo.llama3.icr(cfg, **kwargs)

chatml transforms for datasets with system, input, chosen, rejected ex. https://huggingface.co/datasets/argilla/distilabel-intel-orca-dpo-pairs

intel

prompt_strategies.dpo.llama3.intel(cfg, **kwargs)

For Intel Orca DPO Pairs

ultra

prompt_strategies.dpo.llama3.ultra(cfg, **kwargs)

for ultrafeedback binarized conversations