Large Model Role-Playing Training via Portrait-Guided Conversation Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large models struggle to enhance their role-playing ability, particularly in generating customizable and highly anthropomorphic chat robots with emotional warmth, which is crucial for user interaction.
Innovation Solution
A method for training a large model involves obtaining conversation samples with detailed role portraits, plots, and multiple rounds of conversations. The model predicts conversation sentences by inputting these elements into an initial large model and then trains a target model based on the differences between predicted and sample conversation sentences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional large model training methods are used, then the model can process general language tasks, but it fails to achieve high role-playing ability and emotional warmth in chat robot scenarios
Solution Approach 1:
The patent segments the training process into distinct phases: pre-training on general language data, then fine-tuning on role-specific conversation data with portrait information. This segmentation allows the model to first acquire general language capabilities and then specialize in role-playing, resolving the contradiction between general competence and specialized performance.
Solution Approach 2:
The patent introduces role-specific portrait information (local characteristics) into the training data, allowing different parts of the model to specialize in different role characteristics. This enables the model to maintain general language processing while developing specialized role-playing abilities for specific chat robot personas.
2Reliability
If the model is trained with detailed role portraits and multiple conversation rounds, then the role-playing ability improves, but the training data complexity and processing requirements increase
Solution Approach 1:
The patent performs preliminary processing of role portrait information and conversation data before training, organizing them into structured formats with clear role definitions and conversation templates. This preliminary action simplifies the training process by pre-structuring complex data, reducing the computational burden during actual model training.
Solution Approach 2:
The patent introduces an intermediary layer that processes role portrait information and converts it into structured training data with clear role assignments and conversation patterns. This intermediary structure mediates between the complex raw data and the model, making the training process more manageable while maintaining high conversation accuracy.
3Reliability
If the model generates highly anthropomorphic and emotionally warm responses, then user satisfaction improves, but the computational resources and training time increase
Solution Approach 1:
The patent enables the model to learn emotional warmth and anthropomorphic response patterns through self-supervised learning on role-play conversation data, where the model automatically learns to generate emotionally appropriate responses without requiring extensive manual annotation or complex external systems. This self-service approach reduces training time while maintaining emotional quality.
Data Source
AI summary
The disclosure discloses a method for training a large model, an apparatus for training a large model, an electronic device and a storage medium, and relates to a field of computer technologies, especially to a field of artificial intelligence technologies such as large model and deep learning. The method includes: obtaining a conversation sample, in which the conversation sample includes portraits of a plurality of roles, a plot containing the plurality of roles and a plurality of rounds of conversations among the plurality of roles; for any one of the plurality of roles, obtaining a predicted conversation sentence of the role by inputting the portraits of the plurality of roles, the plot and a historical conversation sentence corresponding to a sample conversation sentence of the role in the plurality of rounds of conversations into an initial large model; and obtaining a target large model by training the initial large model according to a difference between the predicted conversation sentence and the sample conversation sentence.