prompt_strategies.orpo.chat_template

prompt_strategies.orpo.chat_template

chatml prompt tokenization strategy for ORPO

Classes

Name	Description
Message	message/turn
MessageList	conversation
ORPODatasetParsingStrategy	Strategy to parse chosen rejected dataset into messagelist
ORPOPrompter	Single Turn prompter for ORPO
ORPOTokenizingStrategy	rejected_input_ids

Message

prompt_strategies.orpo.chat_template.Message()

message/turn

MessageList

prompt_strategies.orpo.chat_template.MessageList()

conversation

ORPODatasetParsingStrategy

prompt_strategies.orpo.chat_template.ORPODatasetParsingStrategy()

Strategy to parse chosen rejected dataset into messagelist

Methods

Name	Description
get_chosen_conversation_thread	Dataset structure mappings
get_prompt	Map the data to extract everything up to the last turn
get_rejected_conversation_thread	Dataset structure mappings

get_chosen_conversation_thread

prompt_strategies.orpo.chat_template.ORPODatasetParsingStrategy.get_chosen_conversation_thread(
    prompt,
)

Dataset structure mappings

get_prompt

prompt_strategies.orpo.chat_template.ORPODatasetParsingStrategy.get_prompt(
    prompt,
)

Map the data to extract everything up to the last turn

get_rejected_conversation_thread

prompt_strategies.orpo.chat_template.ORPODatasetParsingStrategy.get_rejected_conversation_thread(
    prompt,
)

Dataset structure mappings

ORPOPrompter

prompt_strategies.orpo.chat_template.ORPOPrompter(
    self,
    chat_template,
    tokenizer,
)

Single Turn prompter for ORPO

ORPOTokenizingStrategy

prompt_strategies.orpo.chat_template.ORPOTokenizingStrategy(
    self,
    *args,
    dataset_parser=None,
    **kwargs,
)

rejected_input_ids input_ids rejected_attention_mask attention_mask rejected_labels labels

Functions

Name	Description
load	chatml transforms for datasets with system, input, chosen, rejected

load

prompt_strategies.orpo.chat_template.load(tokenizer, cfg, ds_cfg=None, **kwargs)

chatml transforms for datasets with system, input, chosen, rejected