Skip to content

chat

ChatListWiseReward

Bases: BaseHelpfulnessListWiseReward

Chat: Simulates human conversation and communicates a variety of topics through text understanding and generation, emphasizing coherence and natural flow of interaction.

Source code in rm_gallery/gallery/rm/alignment/helpfulness/chat.py 
22
23
24
25
26
27
28
29
@RewardRegistry.register("chat_listwise_reward")
class ChatListWiseReward(BaseHelpfulnessListWiseReward):
    """Chat: Simulates human conversation and communicates a variety of topics through text understanding and generation, emphasizing coherence and natural flow of interaction."""

    name: str = Field(default="chat_listwise_reward", description="reward name")
    scenario: str = Field(default=SCENARIO, description="assistant scenario")
    principles: List[str] = Field(default=PRINCIPLES)
    desc: str = Field(default=DESC)