Object interaction method and device

Abstract

The invention provides an object interaction method and device. The object interaction method comprises the steps that an interaction instruction submitted by a target object for an auxiliary object in a game scene is determined; in response to the interaction instruction, the auxiliary object information, the target object information and the scene information of the game scene are determined, information related to the interaction instruction is obtained, and it is ensured that the obtained information conforms to the game scene at the interaction instruction triggering moment. And constructing interaction information based on the auxiliary object information, the target object information and the scene information, and generating interaction description information corresponding to the interaction information by using a large language model. The interaction information is processed by utilizing the large language model, the behavior or dialogue for driving the auxiliary object and the target object to perform personalized interaction is predicted, and the auxiliary object and the target object are driven to perform interaction based on the interaction description information, so that the flexibility and authenticity of the behavior decision of the auxiliary object are improved, and the game interaction experience of a game player is improved.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to object interaction methods and devices. Background Art

[0002] With the development of artificial intelligence and the emergence of large language models, such as ChatGPT, players are embracing the concept of NPCs (non-player characters) in games. Players are anticipating more human-like, intelligent NPCs. Existing technologies typically pre-set interaction methods for NPCs in games. During gameplay, NPCs interact with players based on these pre-set interaction methods, explaining fixed plot points and performing pre-set actions. This type of NPC-player interaction results in rigid behavior and a lack of adaptability, resulting in a lack of flexibility and authenticity in their interactions. Therefore, a more effective object interaction method is urgently needed to address these issues. Summary of the Invention

[0003] In view of this, embodiments of this specification provide an object interaction method, an object interaction apparatus, a computing device, a computer-readable storage medium, and a computer program product to solve the above-mentioned problems in the prior art.

[0004] According to a first aspect of an embodiment of this specification, there is provided an object interaction method, including: Determine the interaction instructions submitted by the target object to the auxiliary object in the game scene; determining auxiliary object information, target object information, and scene information of the game scene in response to the interaction instruction, and constructing interaction information based on the auxiliary object information, the target object information, and the scene information; The large language model is used to generate interaction description information corresponding to the interaction information, and the auxiliary object is driven to interact with the target object based on the interaction description information.

[0005] Optionally, in a case where the interaction instruction is a dialogue interaction instruction, determining the auxiliary object information, the target object information, and the scene information of the game scene in response to the interaction instruction includes: collecting auxiliary object attribute information, object memory information, and dialogue style information of the auxiliary object in response to the dialogue interaction instruction, and using the auxiliary object attribute information, the object memory information, and the dialogue style information as the auxiliary object information; In response to the dialogue interaction instruction, target object dialogue information corresponding to the target object is collected, as well as scene information of the game scene, and the target object dialogue information is used as the target object information.

[0006] Optionally, driving the auxiliary object to interact with the target object based on the interaction description information includes: Constructing a dialogue content that matches the dialogue style information based on the interaction description information, and driving the auxiliary object to interact with the target object based on the dialogue content.

[0007] Optionally, in a case where the interaction instruction is a behavior interaction instruction, determining the auxiliary object information, the target object information, and the scene information of the game scene in response to the interaction instruction includes: determining a target object behavior and a target object dialogue of the target object, an auxiliary object behavior and an auxiliary object dialogue of the auxiliary object, and a target context of the game scene in response to the behavior interaction instruction; The target object behavior and the target object dialogue are converted into the target object information, the auxiliary object behavior and the auxiliary object dialogue are converted into the auxiliary object information, and the target situation is converted into the scene information.

[0008] Optionally, driving the auxiliary object to interact with the target object based on the interaction description information includes: Determine interaction behavior parameters based on the interaction description information, and drive the auxiliary object and the target object to complete behavior interaction based on the interaction behavior parameters.

[0009] Optionally, the training of the large language model includes: Determining a pre-trained large language model, and setting the auxiliary object for the game scene; Distributed training is performed on the branch parameters of the pre-trained large language model based on the configuration information of the auxiliary object to obtain a large language model that meets the training stop condition.

[0010] Optionally, the generating interaction description information corresponding to the interaction information by using a large language model includes: Determining key-value intermediate information associated with the large language model; The large language model is used to predict the interaction information based on the key-value intermediate information to obtain the interaction description information corresponding to the interaction information.

[0011] Optionally, determining the key-value intermediate information associated with the large language model includes: Determining initial key-value intermediate information associated with the large language model; The initial key-value intermediate information is divided according to a preset division rule to obtain the key-value intermediate information.

[0012] According to a second aspect of an embodiment of this specification, there is provided an object interaction device, including: A determination module is configured to determine an interaction instruction submitted by a target object to an auxiliary object in a game scene; a construction module configured to determine auxiliary object information, target object information, and scene information of the game scene in response to the interaction instruction, and construct interaction information based on the auxiliary object information, the target object information, and the scene information; A generation module is configured to generate interaction description information corresponding to the interaction information using a large language model, and drive the auxiliary object to interact with the target object based on the interaction description information.

[0013] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising a memory, a processor, and a computer program or instructions stored in the memory and executable on the processor, wherein the processor implements the steps of the object interaction method when executing the computer program or instructions.

[0014] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program or instructions, and when the computer program or instructions are executed by a processor, the steps of the object interaction method are implemented.

[0015] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program or instructions, which implement the steps of the above-mentioned object interaction method when executed by a processor.

[0016] The object interaction method provided in this specification determines the interaction instructions submitted by the target object to the auxiliary object in the game scene. In response to the interaction instructions, the auxiliary object information, the target object information and the scene information of the game scene are determined to obtain information related to the interaction instructions, and ensure that the information obtained conforms to the game scene at the moment the interaction instructions are triggered. The interaction information is constructed based on the auxiliary object information, the target object information and the scene information, and the interaction description information corresponding to the interaction information is generated using a large language model. The interaction information is processed using a large language model to predict the behavior or dialogue that drives the auxiliary object to interact with the target object in a personalized manner, and the auxiliary object is driven to interact with the target object based on the interaction description information, thereby improving the flexibility and authenticity of the auxiliary object's behavior decision-making and improving the game interaction experience of game players. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of an object interaction method provided in one embodiment of this specification; Figure 2 This is a framework diagram of an NPC dialogue system corresponding to an object interaction method provided in an embodiment of this specification; Figure 3 This is a diagram of an intelligent NPC personalized behavior decision framework corresponding to an object interaction method provided in an embodiment of this specification; Figure 4 This is a processing flow chart of an object interaction method applied to game NPC interaction provided by an embodiment of this specification; Figure 5 This is a diagram of an intelligent NPC system architecture based on a large language model corresponding to an object interaction method provided in an embodiment of this specification; Figure 6 This is a schematic diagram of the structure of an object interaction device provided in one embodiment of this specification; Figure 7 This is a structural block diagram of a computing device provided in one embodiment of this specification. DETAILED DESCRIPTION

[0018] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0019] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a," "the," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0020] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0021] In one or more embodiments of this specification, the large language model is a natural language processing technology based on deep learning, which can automatically learn the probability distribution of text data and generate new text that conforms to the distribution. In the large language model, Transformer is widely used in tasks such as language modeling, text generation, and machine translation, and has achieved good results. This is because Transformer can better capture long-distance dependencies in text, and can be calculated in parallel, which makes training faster. In Transformer, the self-attention mechanism is used to calculate the similarity between the representation of each position in the input sequence and the representation of other positions. This similarity can be used to weightedly calculate the contextual representation of each position, thereby better capturing the semantic information in the text. At the same time, Transformer also introduces technologies such as residual connections and layer normalization to accelerate training and improve model performance.

[0022] In large language models, the Attention mechanism is a core component of the Transformer. It helps the model better focus on important parts of the input sequence. In this mechanism, the model calculates the similarity between the input at the current position and other positions, and then performs a weighted sum of the representations of other positions based on the similarity weights to obtain the contextual representation of the current position. This mechanism is widely used in tasks such as text generation and machine translation, helping the model better capture the correspondence between semantic information and language.

[0023] Before training a large language model, the original text needs to be preprocessed, including tokenization, stop word removal, and conversion to a numerical representation. This processing makes the text data more suitable for neural network training. The training goal is to maximize the probability of a given text data, which is achieved by minimizing the cross-entropy loss function. When generating new text, the large language model generates the next word or character based on the existing text content. This process can be achieved through methods such as greedy search, beam search, or sampling.

[0024] Large language models can be further improved through fine-tuning and transfer learning. Fine-tuning involves using new text data to perform supervised fine-tuning on a pre-trained model. Transfer learning involves applying a pre-trained model to different natural language processing tasks, such as machine translation and text classification. These methods can help models better adapt to new tasks and data, improving their generalization and performance.

[0025] In summary, large language models are a powerful natural language processing technology that automatically learns the probability distribution of text data and generates new text that conforms to this distribution. The Transformer and Attention mechanisms are core technologies in large language models, helping them better capture the correspondence between semantic information and language. When training and applying large language models, attention should be paid to preprocessing, training objectives, prediction procedures, fine-tuning, and transfer learning to improve model performance and applicability.

[0026] First, the terms involved in one or more embodiments of this specification are explained.

[0027] NPC (Non-Player Character): In video games, NPCs are characters controlled by the game's programming rather than by the player. These characters play various roles within the game, such as providing quests, posting information, conducting trades, and participating in combat, adding rich storylines and interactivity to the game.

[0028] The Transformer model is a deep learning model based on the self-attention mechanism. It is essentially an encoder-decoder structure. The encoder is responsible for reading and understanding the input text, while the decoder is responsible for generating the output text. They interact with each other through the attention mechanism.

[0029] Attention mechanism: A technology that allows the model to focus on important information and fully learn and absorb it. Its core idea is to calculate the weighted sum of each position through the relationship between query, key, and value.

[0030] LoRA (Low-Rank Adaptation): A fine-tuning technique for large pre-trained models (such as large language models (LLMs)). While keeping the original model parameters unchanged, it adds a small number of trainable parameters to the model. The core idea is to adjust the behavior of the pre-trained model by introducing a small number of trainable parameters without retraining the entire model, significantly reducing the computing resources and time required for training.

[0031] VLLM: An efficient generative AI model inference framework, specifically optimized for inference performance on large models. It aims to provide a fast and scalable solution for deploying large-scale language models (LLMs) in production environments. VLLM supports a variety of advanced technologies to improve inference speed and efficiency, such as the PagedAttention mechanism and KV-Cache.

[0032] PagedAttention: An innovative attention mechanism implementation introduced in VLLM, primarily aimed at improving efficiency when processing long sequences. PagedAttention organizes and accesses this associated data in a new way, maintaining high computational efficiency and low memory usage even when processing very long sequences. This approach allows the model to more efficiently utilize hardware resources, significantly improving throughput and reducing latency, especially when running on GPUs.

[0033] KV-Cache (Key-Value Cache): A technique used to accelerate the computation of the self-attention layer in the Transformer architecture. When generating text, especially word-by-word, KV-Cache can store previously computed key and value tensors. This allows these computational results to be reused when generating the next word, rather than recalculating the self-attention representation for the entire sequence each time. This not only significantly speeds up inference but also reduces resource waste caused by repeated computation.

[0034] Token: In natural language processing (NLP) and Transformer models, a token is the smallest processing unit into which raw text is segmented. It can be a word, a subword, a character, or, in certain scenarios, a special symbol.

[0035] Multi-Head Attention (MHA): One of the core components of the Transformer model, it is used to capture the dependencies between different positions in the input sequence. It allows the model to focus on information in different representation subspaces by performing multiple attention mechanisms in parallel (each called a "head").

[0036] Multi-Query Attention: MQA is a simplified form of the standard Multi-HeadAttention (MHA). It reduces computation and memory overhead by sharing key and value vectors.

[0037] Grouped-Query Attention: A compromise between MHA and MQA. It groups multiple query headers, with each group sharing a set of keys and values.

[0038] In this specification, an object interaction method is provided. This specification also relates to an object interaction apparatus, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0039] Figure 1 A flowchart of an object interaction method provided according to an embodiment of this specification is shown, which specifically includes the following steps: Step 102: Determine the interaction instruction submitted by the target object to the auxiliary object in the game scene.

[0040] Specifically, the target object can be a game character representing the player in the game scene. Players can play the game through a terminal device and manipulate the character to perform game tasks within the game scene. The game scene is the game scene generated after entering the game through a terminal device. In the game scene, the character typically needs to perform game tasks to earn rewards. Game tasks include, but are not limited to, completing levels and battling other players. An auxiliary object refers to a non-player character (NPC) in the game scene. An auxiliary object can be a character NPC, answering questions raised by the player during the game, engaging in dialogue with the character, making behavioral decisions based on the player's behavior and dialogue with the player, and interacting with the character. An auxiliary object can also be a monster NPC, assisting the character in performing game tasks or acting as an opponent for the character. The player must defeat the auxiliary object to complete the game task. An interactive instruction is a computer instruction triggered by the target object through contact, dialogue, or other interactions with the auxiliary object. An interactive instruction triggers the auxiliary object to interact with the target object or make behavioral decisions.

[0041] Based on this, in the game scene, the target object can interact with the auxiliary object. When the target object interacts with the auxiliary object, the interaction command for the auxiliary object is triggered. The auxiliary object interacts with the target object based on the interaction command, including but not limited to answering questions raised by the target object and making behavioral decisions based on the interaction command, thus realizing the behavioral interaction between the auxiliary object and the target object.

[0042] Step 104: Determine auxiliary object information, target object information, and scene information of the game scene in response to the interaction instruction, and construct interaction information based on the auxiliary object information, the target object information, and the scene information.

[0043] Specifically, after determining the interaction instructions submitted by the target object to the auxiliary object in the game scene, the auxiliary object information, target object information, and scene information of the game scene are determined in response to the interaction instructions, and interaction information is constructed based on the auxiliary object information, target object information, and scene information. The auxiliary object information includes but is not limited to the dialogue information and behavioral interaction information generated by the auxiliary object through interaction with the target object in the game scene, the auxiliary object's historical dialogue information and historical interaction information, as well as the auxiliary object's attribute information and the auxiliary object's record of game environment knowledge. The auxiliary object's attribute information can be basic information settings such as the auxiliary object's name, age, dialogue style, and behavioral memory; the game environment knowledge can be knowledge stored in the knowledge base of the game world corresponding to the game scene. The target object information includes but is not limited to the target object's behavioral description information and the dialogue information between the target object and the auxiliary object; the scene information of the game scene refers to the game context corresponding to the game scene; the interaction information is an integrated summary of the auxiliary object information, target object information, and scene information, and is used to drive the interaction between the auxiliary object and the target object by analyzing the interaction information.

[0044] Based on this, after determining the interaction instructions submitted by the target object to the auxiliary object in the game scene, in response to the interaction instructions, auxiliary object information related to the auxiliary object, target object information related to the target object, and scene information of the game scene are collected, and the auxiliary object information, target object information and scene information are integrated to construct interaction information. The interaction information is used to make behavioral decisions and / or dialogue decisions for the auxiliary object.

[0045] Step 106: Generate interaction description information corresponding to the interaction information using a large language model, and drive the auxiliary object to interact with the target object based on the interaction description information.

[0046] Specifically, after determining the auxiliary object information, target object information and scene information of the game scene in response to the interaction instruction, and constructing the interaction information based on the auxiliary object information, target object information and scene information, the large language model can be used to generate interaction description information corresponding to the interaction information, and the auxiliary object and the target object can be driven to interact based on the interaction description information, wherein the interaction description information includes a text description of the game scene, a text description of the target object behavior, and a text description of the auxiliary object behavior.

[0047] Based on this, after the auxiliary object information, target object information and scene information of the game scene are determined in response to the interaction instruction, and the interaction information is constructed based on the auxiliary object information, target object information and scene information, the interaction information is input into the textual processing of the game scene, target object behavior and auxiliary object behavior using the large language model, and the game scene, target object behavior and auxiliary object behavior are respectively converted into textual description content, that is, interaction description information, and then the auxiliary object is driven to conduct dialogue interaction or behavior interaction with the target object based on the generated interaction description information.

[0048] In practical applications, the dialogue interaction between the auxiliary object (user-game player) and the target object can be achieved through Figure 2 The NPC dialogue system framework is shown in the figure. This framework uses a large language model as the core of NPC corpus generation. Leveraging the emerging intelligence of the large language model, it generates natural, fluent, and personalized dialogue responses. NPCs are typically created in game scenes. Their character and memory are loaded based on their design and the interaction memories between the NPC and the corresponding game character. Questions posed by the user (game character) are determined, and context information is collected. Using an embedding module, the NPC's character, interaction memories, user questions, and context information are vectorized, summarizing the most recent interaction memories with the user. These interaction memories are then organized using the large language model. The final prompt content is obtained by integrating the user question and context information, along with the character information and interaction memories. The large language model is then used to process the prompt content and the interaction memories between the NPC and the game character to generate NPC responses with a unique language style, enabling intelligent, natural, fluent, and personalized NPC dialogue.

[0049] The behavioral interaction between the auxiliary object (user - game player) and the target object can be achieved through Figure 3 The illustrated framework implements a personalized intelligent NPC behavior decision-making framework. NPC behavior decision-making is achieved by collecting game scene environmental information, NPC memories, and interaction information between NPCs and players, and processing this information using a large language model. In specific implementation, the collected environmental information, NPC memories, and interaction information between NPCs and players are formatted and input into the large language model for autonomous intelligent decision-making. NPC decision information is then formatted and parsed to determine behavior decisions such as NPC speaking, NPC walking, and NPC picking up. Based on the NPC's behavioral decisions, the NPC interacts with game characters, achieving more intelligent NPC decision-making.

[0050] Furthermore, considering that the interaction between the target object and the auxiliary object can be a dialogue interaction, when the interaction instruction is a dialogue interaction instruction, the dialogue between the target object and the auxiliary object can be generated by collecting the auxiliary object information of the auxiliary object, the target object information of the target object, and the scene information of the game scene, and performing information analysis. The specific implementation is as follows: Step 106-2: collecting auxiliary object attribute information, object memory information, and dialogue style information of the auxiliary object in response to the dialogue interaction instruction, and using the auxiliary object attribute information, the object memory information, and the dialogue style information as the auxiliary object information; Step 106 - 4 : In response to the dialogue interaction instruction, target object dialogue information corresponding to the target object is collected, as well as scene information of the game scene, and the target object dialogue information is used as the target object information.

[0051] Specifically, auxiliary object attribute information may include basic information settings such as the auxiliary object's name, age, species, character, dialogue style, and memory. Object memory information refers to the auxiliary object's stored behavioral memory, dialogue memory, and scene memory. Dialogue style information refers to the auxiliary object's dialogue style, which can be humorous, aloof, or cute. The target object's corresponding target object dialogue information may include questions posed by the target object to the auxiliary object, requiring the auxiliary object to answer and provide the corresponding target object dialogue information.

[0052] Based on this, when the interaction instruction is a dialogue interaction instruction, it means that the target object has raised a question to be answered to the auxiliary object, and the auxiliary object needs to answer the question to be answered. When answering the question, it is necessary to refer to the auxiliary object's auxiliary object attribute information, object memory information, and object style information. In response to the dialogue interaction instruction, the auxiliary object's auxiliary object attribute information, object memory information, and dialogue style information are collected, and the auxiliary object attribute information, object memory information, and dialogue style information are used as the auxiliary object information. In response to the dialogue interaction instruction, the target object dialogue information corresponding to the target object is collected, and the game scene is determined, the game scene is described in text to generate scene information of the game scene, and the target object dialogue information is used as the target object information. It is used for subsequent dialogue interaction between the target object and the auxiliary object.

[0053] For example, a game scene includes a target object (game character) corresponding to the player and an auxiliary object (NPC). The game character is controlled by the player and can interact with the NPC. During the dialogue interaction between the game character and the NPC, the NPC can answer the game character's questions and engage in conversation with the game character within the game scene. When the game character asks the NPC a question, a dialogue interaction command is triggered. In response to the dialogue interaction command, the game server collects information: dialogue information (questions to be answered) of the game character (target object) and attribute information, memory information, and conversation style information of the NPC. NPC attribute information can include the NPC's name, species, and self-proclaimed name; memory information can include recent behavioral memories (e.g., recently completed tasks) and conversation style information (e.g., aloof, cute, enthusiastic, etc.). Furthermore, a textual description of the game scene is generated to generate scene information. This allows for the generation of personalized NPC dialogue responses.

[0054] To sum up, when the interaction instruction is a dialogue interaction instruction, the dialogue interaction can be completed by collecting the auxiliary object information of the auxiliary object, the target object information of the target object, and the scene information of the game scene, and generating a dialogue between the target object and the auxiliary object through comprehensive analysis of the information.

[0055] Furthermore, in the case where the interaction instruction is a dialogue interaction instruction, after determining the interaction description information, the dialogue content that the auxiliary object can provide to the target object can be constructed based on the interaction description information, and the dialogue content conforms to the dialogue style of the auxiliary object. The specific implementation is as follows: Step 106-4-2: Constructing a dialogue content that matches the dialogue style information based on the interaction description information, and driving the auxiliary object to interact with the target object based on the dialogue content.

[0056] Specifically, the conversation content matches the conversation style information of the auxiliary object, which is the text content that the auxiliary object will tell the target object. The conversation content can be displayed to the target object in the form of text or played in the form of audio to tell the target object.

[0057] Based on this, if the interaction instruction is a dialogue interaction instruction, after determining the interaction description information, dialogue content can be constructed based on the interaction description information to match the dialogue style information, so that the dialogue content matches the language style of the auxiliary object. Based on the dialogue content, the auxiliary object is driven to interact with the target object, and the dialogue content is displayed to the target object through the dialogue interface or played back in the form of voice.

[0058] By analyzing players' unanswered questions, changes in the game environment, NPC attributes (personalities), historical dialogue, historical behavior, and memory, we generate coherent and logical responses. Using the open-source Big Language Model as a foundation, we developed an NPC personality configuration tool based on the Big Language Model. This tool allows a common Big Language Model to simulate multiple different NPCs simply by changing the NPC's personality settings.

[0059] Continuing with the above example, the historical dialogue between the NPC and the game character could be "Player: Do you know anything about the Obsidian Scepter? Ivan (NPC): It is a relic from ancient times. It is said that whoever wields it can control storms. But it disappeared in the Twilight War. Player: Where did the Twilight War take place? Ivan: The battle took place in Shadowdale, where the armies of light and shadow clashed fiercely, causing the valley to be shrouded in fog forever." When the current player's question is "Where do you think the Obsidian Scepter might be now?", the NPC can give a logical answer based on the historical dialogue: "If you want to find the Obsidian Scepter, the Misty Forest in Shadowdale is the starting point. It is said that after the battle, a shadow knight took it deep into the forest, trying to hide its tracks with fog."

[0060] In summary, when the interaction instructions are dialogue interaction instructions, dialogue content that matches the dialogue style information is constructed based on the interaction description information. Based on the dialogue content, the auxiliary object and the target object are driven to engage in dialogue interaction, achieving a natural and smooth dialogue between the auxiliary object and the target object.

[0061] Furthermore, considering that the interaction between the target object and the auxiliary object can be a behavior (action) interaction, when the interaction instruction is a behavior interaction instruction, the auxiliary object information of the auxiliary object, the target object information, and the scene information of the game scene can be collected, and the information can be analyzed to generate the actions that the auxiliary object can perform, and the action interaction can be performed with the target object. The specific implementation is as follows: Step 106-6: determining the target object behavior and target object dialogue of the target object, the auxiliary object behavior and auxiliary object dialogue of the auxiliary object, and the target context of the game scene in response to the behavior interaction instruction; Step 106-8: Convert the target object behavior and the target object dialogue into the target object information, convert the auxiliary object behavior and the auxiliary object dialogue into the auxiliary object information, and convert the target situation into the scene information.

[0062] Specifically, the target object's target behavior includes, but is not limited to, walking, flying, climbing, fighting, and clearing levels within the game scene, as well as cooperating with or competing with auxiliary objects. The target object's target object dialogue refers to the speech generated by the target object within the game scene, including, but not limited to, the target object's soliloquy, dialogues between the target object and auxiliary objects, and declarations. The auxiliary object behavior of the auxiliary object refers to the auxiliary object's historical behavior within the game, including, but not limited to, assisting the target object and intercepting the target object. The auxiliary object dialogue refers to the dialogue between the auxiliary object and the target object within the game, or the plot text that introduces the game plot. This plot text can be displayed in the game interface in text form or played as audio to advance the game plot. The target context of the game scene refers to the game context. The target context can be stored as a context image or context fragment. When using the target context, the target context can be described as text content, i.e., context information. By describing the target object behavior and combining it with the target object dialogue, target object information can be obtained.

[0063] Based on this, when the interaction instruction is a behavioral interaction instruction, the target object behavior and target object dialogue of the target object are determined in response to the behavioral interaction instruction, the auxiliary object behavior and auxiliary object dialogue of the auxiliary object are determined, and the target context of the game scene is determined. The auxiliary object behavior can be determined based on the auxiliary object's memory, which can include the auxiliary object's character settings and historical behavior. The target object behavior is described and converted into textual behavior description information. The behavior description information and the target object dialogue are integrated to obtain target object information. The auxiliary object behavior is described and converted into textual description information, which is then integrated with the auxiliary object dialogue to obtain auxiliary object information. The target context is also described and converted into scene information.

[0064] Continuing with the previous example, the game scene includes the target object (game character) corresponding to the player and auxiliary objects (NPCs). The game character is controlled by the player and can interact with the NPC. When the game character and NPC interact, behavior and dialogue can be collected for each. By describing the collected behaviors, they can be converted into behavioral descriptions. The game character's behavior is the target object's behavior. By describing the target object's behavior, the game character's behavior description is obtained. The game character's behavior description includes, but is not limited to, the game character's position in the game scene, the status of resources such as health, and combat status. The NPC's behavior is the auxiliary object's behavior. By describing the auxiliary object's behavior, the NPC's behavior description is obtained. The NPC's behavior description includes, but is not limited to, the NPC's position in the game scene, the status of resources such as health, and combat status. Analyze the game context to determine the behavioral feedback the NPC can provide to the game character. When converting the game context into text, a text template can be used: "Current health xxx, current location xxx, current status xxx." By identifying the game context and filling in the text template, the scene information is obtained.

[0065] To sum up, by collecting the auxiliary object information of the auxiliary object, the target object information of the target object, and the scene information of the game scene, and performing information analysis to generate the actions that the auxiliary object can perform, and interact with the target object, the accuracy and authenticity of the auxiliary object's behavior decision-making can be improved.

[0066] Furthermore, in the case where the interaction instruction is a behavior interaction instruction, after the interaction description information is determined, the interaction behavior parameters can be determined based on the interaction description information, which is specifically implemented as follows: Step 106-8-2: Determine interaction behavior parameters based on the interaction description information, and drive the auxiliary object and the target object to complete behavior interaction based on the interaction behavior parameters.

[0067] Specifically, interaction behavior parameters refer to action parameters, representing the range, distance, speed, movement, attack, skill, expression, interaction, and defense of the interaction between the assist and target objects. Interaction behavior parameters represent values, state parameters, code instructions, and physical properties.

[0068] Based on this, the interaction behavior parameters are determined based on the interaction description information, and the auxiliary object and the target object are driven to complete the behavioral interaction based on the interaction behavior parameters, so as to realize the intelligent decision-making of the auxiliary object and improve the continuity and rationality of the game connection.

[0069] With the above example, the interactive behavior parameter can be used to drive the interaction between the NPC and the game character, and can define the action range of the interaction between the NPC and the game character. The interaction between the NPC and the game character includes but is not limited to action interaction, language interaction and expression interaction. When the game character and the NPC interact, the NPC can fight or cooperate with the game character. In the case of hostile relationship between the NPC and the game character, the NPC can resist and counterattack according to the game character's move. In the case of cooperative relationship between the NPC and the game character, the NPC will cooperate with the game character's behavior, including cooperation with the game character, joint defense against enemy attack, and completion of game task. The interaction description information is parsed in the form of keyword matching, such as the action of shooting. When the keyword "shooting" is obtained by parsing the interaction description information, the corresponding shooting module of the NPC is called to execute the shooting action. The NPC has a predetermined task to move along a route, but in the process of moving forward, it sees a material, and the NPC may make a behavior decision to pick up the material instead of passing by the material and moving along the fixed route.

[0070] In summary, the interactive behavior parameter is determined based on the interaction description information, and the auxiliary object and the target object complete the behavior interaction based on the interactive behavior parameter, realize the intelligent decision of the auxiliary object, improve the continuity and logicality of the game connection, and improve the game experience of the game player.

[0071] Further, considering that the pre-trained large language model cannot meet the individualized needs in the game scene, therefore, the model fine-tuning can be performed based on the pre-trained large language model and the configuration information of the auxiliary object, to realize the purpose of simulating different auxiliary objects by a general large language model, and the specific implementation is as follows: Step 106-10: determining a pre-trained large language model, and setting the auxiliary object for the game scene; Step 106-12: based on the configuration information of the auxiliary object, performing distributed training on the branch parameters of the pre-trained large language model, to obtain a large language model meeting the training stop condition.

[0072] Specifically, the pre-trained large language model can be a pre-trained Transformer model. The configuration information of the auxiliary object can be the character setting information of the auxiliary object. For example, the auxiliary object's name is A, its species is monkey, it calls itself B, and its most recent memory is returning from a trip. The branch parameters of the large language model correspond to the trainable layers (network layers) injected into the Transformer module in the pre-trained large language model. When fine-tuning the pre-trained large language based on the configuration information, only the network layer parameters can be trained to increase the training speed. The training stopping condition can be that a preset training round is reached, the model prediction accuracy reaches a preset accuracy threshold, or the model fine-tuning training reaches a preset training time.

[0073] Based on this, a pre-trained large language model and an auxiliary object for the game scenario are determined. Configuration information such as the auxiliary object's name, species, and memories is obtained. A network layer is added to the pre-trained large language model. Distributed training is then performed on the network layer parameters (branch parameters) of the newly added network layer in the pre-trained large language model based on this configuration information until a large language model is obtained that meets the training stop criteria and is capable of generating conversational content and behavioral descriptions that match the auxiliary object's current personality.

[0074] Continuing with the above example, LoRA incremental learning technology can be used to integrate the relevant data and knowledge of newly created NPCs into the pre-trained large language model, achieving a better sense of playability for the new characters in the game scene. LoRA achieves fine-tuning by freezing the weights of the pre-trained model and injecting a trainable layer (called a rank-factorized matrix) into each Transformer block. The advantage of this approach is that LoRA only trains the parameters of the newly added network layer, rather than the parameters of the entire model, resulting in faster training and reduced computational resources. The pre-trained large language model has a very small intrinsic dimension, meaning that it has a very low-dimensional parameter space. Fine-tuning it can achieve the same effect as fine-tuning in the full parameter space. Furthermore, larger large language models have smaller intrinsic dimensions. Therefore, after model pre-training, parameter updates can be represented by a low-rank factorized matrix, allowing parameters in the newly added branch to be trained instead of the entire model.

[0075] In summary, pre-training the branch parameters of the large language model based on the configuration information of the auxiliary objects for distributed training can improve the fine-tuning training speed of the pre-trained large language model, making it easier for the large language model to adapt to the auxiliary objects in the game scene.

[0076] Furthermore, when using a large language model to process interactive information, KV-Cache technology can be used to accelerate the reasoning process of the large language model. The specific implementation is as follows: Step 106-14: Determine the key-value intermediate information associated with the large language model; Step 106-16: Using the large language model, predict the interaction information based on the key-value intermediate information to obtain the interaction description information corresponding to the interaction information.

[0077] Specifically, key-value intermediate information refers to the key and value information stored in the cache when using KV-Cache technology to accelerate the inference process of large language models. This key-value intermediate information can be used when calculating the next token.

[0078] Based on this, the key-value intermediate information generated by the large language model before calculating the interaction information is determined. Using the large language model, the interaction information is predicted based on the key-value intermediate information to obtain the interaction description information corresponding to the interaction information, thereby reducing the amount of calculation.

[0079] In practical applications, KV-Cache is a technology used in modern LLMs to accelerate model inference. It can effectively control the use of video memory, especially when processing long text inputs. KV-Cache is a storage mechanism that saves previously calculated keys and values ​​for subsequent use, thus avoiding repeated calculations and significantly reducing the amount of calculation and video memory usage. KV-Cache works by first calculating the complete attention and then saving the keys and values ​​in KV-Cache. When calculating the next token, only the query of the current token needs to be calculated and interacted with the previously saved keys and values, thus reducing duplication of work.

[0080] In summary, by using a large language model, interaction information is predicted based on key-value intermediate information to obtain interaction description information corresponding to the interaction information, thereby reducing the amount of calculation and speeding up the calculation of the large language model.

[0081] Furthermore, considering the large amount of information in the key-value intermediate information, when using a large language model to process the interactive information, it is also necessary to consider the memory usage of the initial key-value intermediate information. The specific implementation is as follows: Step 106-18: Determine initial key-value intermediate information associated with the large language model; Step 106-20: Divide the initial key-value intermediate information according to a preset division rule to obtain the key-value intermediate information.

[0082] In practical applications, initial key-value intermediate information refers to key-value information for which storage space has not yet been allocated. Preset partitioning rules can include adjusting the number of heads in Multi-Query-Attention or Grouped-Query-Attention, shortening the length of stored content, and quantizing the width of key-value information. This reduces video memory and main memory usage while maintaining a certain level of accuracy, improving the throughput of large language models and reducing the inference cost of a single request.

[0083] The following combined Figure 4 , taking the application of the object interaction method provided in this specification in the game NPC interaction as an example, the object interaction method is further explained. Figure 4 The following is a processing flow chart of an object interaction method for game NPC interaction provided by an embodiment of this specification, which specifically includes the following steps: Step 402: Determine a pre-trained large language model and auxiliary objects set for the game scenario.

[0084] The Big Language Model (LLM) is a natural language processing technology based on deep learning. It automatically learns the probability distribution of text data and generates new text that conforms to this distribution. The Transformer architecture is a common technique used in LLM technology. It has been widely used in tasks such as language modeling, text generation, and machine translation, achieving excellent results. This is because, compared to traditional natural language processing techniques, the Transformer can better capture long-range dependencies in text and can be computed in parallel, making training faster. In the Transformer, the self-attention mechanism calculates the similarity between the representation of each position in the input sequence and the representations of other positions. This similarity is used to weight the contextual representation of each position, thereby better capturing the semantic information in the text. Furthermore, the Transformer introduces techniques such as residual connections and layer normalization to accelerate training and improve model performance.

[0085] In large language models, the Attention mechanism is a core component of the Transformer. It helps the model better focus on important parts of the input sequence. In this mechanism, the model calculates the similarity between the input at the current position and other positions, and then performs a weighted sum of the representations of other positions based on the similarity weights to obtain the contextual representation of the current position. This mechanism is widely used in tasks such as text generation and machine translation, helping the model better capture the correspondence between semantic information and language.

[0086] Before training a large language model, the original text needs to be preprocessed, including tokenization, stop word removal, and conversion to a numerical representation. This processing makes the text data more suitable for neural network training. The training goal is to maximize the probability of the next text data given a given text, while minimizing the cross-entropy loss function. When generating new text, the large language model generates probabilities for all words or characters in the vocabulary based on the existing text content and selects the one with the highest probability as the next word.

[0087] Large language models can be further improved through fine-tuning and transfer learning. Fine-tuning involves using new text data to perform supervised fine-tuning on a pre-trained model. Transfer learning involves applying a pre-trained model to different natural language processing tasks, such as machine translation and text classification. These methods can help models better adapt to new tasks and data, improving their generalization and performance.

[0088] In this embodiment, when adding new NPCs (helper objects) to the game, LoRA fine-tuning technology can be used to inject trainable parameters into the original transformer structure. During incremental learning, only these parameters are trained, thus solving the problem of excessive parameters in the entire model that makes training difficult. Furthermore, distributed training technology is used during training to speed up training.

[0089] Step 404: Distributed training is performed on the branch parameters of the pre-trained large language model using the configuration information of the auxiliary object to obtain a trained large language model.

[0090] In practice, LoRA incremental learning technology is used to integrate the relevant data and knowledge of newly created NPCs into the large language model, achieving a better sense of playability for the new characters. LoRA fine-tuning technology addresses the difficulty of large language models in simulating new characters that have not appeared in the training data. This also reduces the number of parameters required to train the large language model and accelerates training through distributed training.

[0091] Step 406: Determine the interaction instruction submitted by the target object to the auxiliary object in the game scene, and determine the auxiliary object information, the target object information and the scene information of the game scene in response to the interaction instruction.

[0092] Target objects refer to the characters representing the player in the game scene, while auxiliary objects refer to the NPCs that interact with the player in the scene, such as through dialogue, combat, and assisting in completing tasks. Players can trigger interactions with NPCs through various means, such as approaching or touching them, generating interaction commands. When an interaction command is triggered, the game server collects information about the player, NPC, and game scene at that moment to construct the current game environment. Auxiliary object information refers to information collected about the NPC after the player triggers an interaction with the NPC. Target object information refers to information collected about the player after the player triggers an interaction with the NPC.

[0093] In a specific implementation, if the interaction instruction is a dialogue interaction instruction, auxiliary object attribute information, object memory information, and dialogue style information of the auxiliary object are collected in response to the dialogue interaction instruction, and the auxiliary object attribute information, object memory information, and dialogue style information are used as the auxiliary object information collected for the auxiliary object. Furthermore, target object dialogue information corresponding to the target object is collected in response to the dialogue interaction instruction, as well as scene information of the game scene, and the target object dialogue information is used as the target object information collected for the target object.

[0094] In practical applications, when the interactive command is a dialogue command, it indicates that the game player needs to engage in a conversation with an NPC. When the game player asks a question, the game player's behavior information, historical conversation information, as well as the NPC's historical conversation information, memory information, conversation style information, and basic settings (name, species, and other personalities) can be collected. Perceivable scene information such as the map location in the game environment is also required. The NPC's response to the player's question can then be generated based on this collected information.

[0095] If the interaction instruction is a behavioral interaction instruction, it indicates that behavioral interaction is required between the game player and the NPC. In this case, the target object behavior and target object dialogue of the target object, the auxiliary object behavior and auxiliary object dialogue of the auxiliary object, and the target context of the game scene are determined in response to the interaction instruction, thereby collecting the game player behavior, NPC behavior, and game context. The collected target object behavior and target object dialogue are converted into target object information, the auxiliary object behavior and auxiliary object dialogue are converted into auxiliary object information, and the target context is converted into scene information.

[0096] In actual applications, the NPC environment context literalization tool can be used to convert the environment in which the NPC is located, the historical dialogue information of the player, and the experience of the NPC into a literal description. The NPC environment context literalization tool can be a large language model. The large language model can be used to describe the environment in which the NPC is located and generate a literal description of the environment in which the NPC is located. The large language model can be used to organize the historical dialogue information between the player and the NPC and generate dialogue content in the form of text. The large language model can also convert the experience of the NPC into a literal description with a clear timeline. Subsequently, the NPC can be guided to make behavior decisions and interact with the game player based on the literal description.

[0097] In addition, the NPC environment context literalization tool can be a combination of a pre-established literal template and a large language model. The literal template can be supplemented and improved based on the environment in which the NPC is located, the historical dialogue information of the player, and the experience of the NPC. For example, the literal template can be: "NPC is located at xxx, NPC's speech content xxx, NPC's historical behavior xxx". After supplementing and improving the literal template, the literal template is input into the large language model for analysis, and the large language model outputs a literal description with coherence and logic.

[0098] Step 408: constructing interaction information based on auxiliary object information, target object information, and scene information.

[0099] Based on the collected information, the input of the large language model is constructed and input into the large language model for processing.

[0100] Step 410: determining initial key-value intermediate information associated with the large language model, and dividing the initial key-value intermediate information according to a preset division rule to obtain key-value intermediate information.

[0101] Step 412: using the large language model to predict the interaction information based on the key-value intermediate information to obtain interaction description information corresponding to the interaction information.

[0102] In practical applications, to address the high inference cost of large language models, KV-Cache technology is employed during the model's execution. This allows previously generated content to be leveraged during inference, reducing memory and video memory usage. The PagedAttention method proposed by the VLLM framework is also employed to significantly improve the throughput of large language models. Specifically, when using a large language model for inference based on interaction information, KV-Cache and the PagedAttention technology proposed by the VLLM framework are employed to increase memory and video memory efficiency, significantly improve the throughput of the large language model, and reduce the inference cost of a single request. The large language model can output interaction description information to guide interactions between NPCs and game players.

[0103] Step 414: driving the auxiliary object to interact with the target object based on the interaction description information.

[0104] In practical applications, if the interaction instructions are dialogue instructions, dialogue content matching the dialogue style is constructed based on the interaction description information. Based on the dialogue content, auxiliary objects are driven to interact with the target object, achieving personalized dialogue between the game player and the NPC. If the interaction instructions are behavioral instructions, indicating that a behavioral interaction is required between the game player and the NPC, the interaction behavior parameters are determined based on the interaction description information, and based on the interaction behavior parameters, the auxiliary object is driven to complete the behavioral interaction with the target object.

[0105] like Figure 5 As shown in the figure, the intelligent NPC system based on the large language model consists of three subsystems: an NPC dialogue system based on the large language model, an intelligent NPC personalized behavior decision system, and a personalized large language model training and reasoning system. The NPC dialogue system based on the large language model includes a personalized NPC response module, an NPC character configuration tool module, and a personalized large language model module; the intelligent NPC personalized behavior decision system includes a patterned decision parsing module, a large language model decision module, and environmental context formatting; and the personalized large language model training and reasoning system includes PagedAttention, LoRA fine-tuning technology, and KV-Cache technology.

[0106] In the NPC dialogue system based on a large language model, the large language model is introduced as the core of the NPC's corpus generation. Leveraging the emerging intelligent capabilities of the large language model, it generates natural, fluent, and personalized dialogue responses. Furthermore, through technologies such as LoRA fine-tuning, the large language model's difficulty in simulating new characters not present in the training data is resolved. Furthermore, the NPC character configuration tool enables the simulation of different NPC speech styles simply through character configuration, achieving an intelligent, natural, fluent, and personalized NPC dialogue system.

[0107] The intelligent NPC personalized behavior decision-making system uses a large language model as the decision-making foundation, enabling intelligent NPC decision-making. Furthermore, to enhance NPC perception, a tool for textualizing NPC environmental context has been developed. This tool converts the NPC's environment, historical conversations with players, and the NPC's own experiences into text and feeds it into the large language model. The large language model then analyzes and judges this environmental information, generating a decision description. Decision information formatting and parsing tools enable the concrete execution of NPC decisions.

[0108] In personalized large language model training and inference systems, the high number of parameters in large language models, as well as the high memory and graphics memory usage that make training difficult and inference costly, are addressed. LoRA fine-tuning technology was introduced to significantly reduce the number of parameters required for training large language models, and distributed training was used to accelerate training. This project also employed the PagedAttention technology proposed by the KV-Cache and VLLM frameworks to increase memory and graphics memory usage efficiency, significantly improve the throughput of large language models, and reduce the inference cost of a single request.

[0109] The object interaction method provided in this specification automatically adjusts the behavior and dialogue of NPCs according to changes in the player's behavior and the game environment to provide a more personalized and realistic game experience. It also reduces the workload of NPC writing and debugging, and improves game development efficiency. In addition, when training large language models, distributed training technology is used to disperse training data to multiple computing nodes for parallel processing, thereby improving the efficiency and speed of model training. By adopting incremental learning technology, the model can be updated and adjusted without retraining the entire model, thereby improving the efficiency and performance of personalized training of the model. By using memory reuse technology, when generating content, tokens in previously generated content are used to reduce the memory and video memory usage in the current generation process.

[0110] Corresponding to the above method embodiment, this specification also provides an object interaction device embodiment, Figure 6 FIG. 1 shows a schematic diagram of the structure of an object interaction device provided by an embodiment of this specification. Figure 6 As shown, the device includes: A determination module 602 is configured to determine an interaction instruction submitted by a target object to an auxiliary object in a game scene; A construction module 604 is configured to determine auxiliary object information, target object information, and scene information of the game scene in response to the interaction instruction, and construct interaction information based on the auxiliary object information, the target object information, and the scene information; The generation module 606 is configured to generate interaction description information corresponding to the interaction information using a large language model, and drive the auxiliary object to interact with the target object based on the interaction description information.

[0111] In an optional embodiment, the building module 604 is further configured to: collecting auxiliary object attribute information, object memory information, and dialogue style information of the auxiliary object in response to the dialogue interaction instruction, and using the auxiliary object attribute information, the object memory information, and the dialogue style information as the auxiliary object information; In response to the dialogue interaction instruction, target object dialogue information corresponding to the target object is collected, as well as scene information of the game scene, and the target object dialogue information is used as the target object information.

[0112] In an optional embodiment, the building module 604 is further configured to: Constructing a dialogue content that matches the dialogue style information based on the interaction description information, and driving the auxiliary object to interact with the target object based on the dialogue content.

[0113] In an optional embodiment, the building module 604 is further configured to: determining a target object behavior and a target object dialogue of the target object, an auxiliary object behavior and an auxiliary object dialogue of the auxiliary object, and a target context of the game scene in response to the behavior interaction instruction; The target object behavior and the target object dialogue are converted into the target object information, the auxiliary object behavior and the auxiliary object dialogue are converted into the auxiliary object information, and the target situation is converted into the scene information.

[0114] In an optional embodiment, the building module 604 is further configured to: Determine interaction behavior parameters based on the interaction description information, and drive the auxiliary object and the target object to complete behavior interaction based on the interaction behavior parameters.

[0115] In an optional embodiment, the generating module 606 is further configured to: Determining a pre-trained large language model, and setting the auxiliary object for the game scene; Distributed training is performed on the branch parameters of the pre-trained large language model based on the configuration information of the auxiliary object to obtain a large language model that meets the training stop condition.

[0116] In an optional embodiment, the generating module 606 is further configured to: Determining key-value intermediate information associated with the large language model; The large language model is used to predict the interaction information based on the key-value intermediate information to obtain the interaction description information corresponding to the interaction information.

[0117] In an optional embodiment, the generating module 606 is further configured to: Determining initial key-value intermediate information associated with the large language model; The initial key-value intermediate information is divided according to a preset division rule to obtain the key-value intermediate information.

[0118] The object interaction method provided in this specification determines the interaction instructions submitted by the target object to the auxiliary object in the game scene. In response to the interaction instructions, the auxiliary object information, the target object information and the scene information of the game scene are determined to obtain information related to the interaction instructions, and ensure that the information obtained conforms to the game scene at the moment the interaction instructions are triggered. The interaction information is constructed based on the auxiliary object information, the target object information and the scene information, and the interaction description information corresponding to the interaction information is generated using a large language model. The interaction information is processed using a large language model to predict the behavior or dialogue that drives the auxiliary object to interact with the target object in a personalized manner, and the auxiliary object is driven to interact with the target object based on the interaction description information, thereby improving the flexibility and authenticity of the auxiliary object's behavior decision-making and improving the game interaction experience of game players.

[0119] The above is a schematic scheme of an object interaction device of this embodiment. It should be noted that the technical scheme of the object interaction device and the technical scheme of the above-mentioned object interaction method belong to the same concept. For details not described in detail in the technical scheme of the object interaction device, please refer to the description of the technical scheme of the above-mentioned object interaction method.

[0120] Figure 7 7 shows a block diagram of a computing device 700 according to an embodiment of the present disclosure. Components of the computing device 700 include, but are not limited to, a memory 710 and a processor 720. The processor 720 is connected to the memory 710 via a bus 730, and a database 750 is used to store data.

[0121] Computing device 700 also includes an access device 740 that enables computing device 700 to communicate via one or more networks 760. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. Access device 740 may include one or more of any type of network interface (e.g., a network interface controller (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0122] In one embodiment of the present specification, the above components of the computing device 700 and Figure 7 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 7 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0123] Computing device 700 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, personal digital assistant, laptop computer, notebook computer, netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or personal computer (PC). Computing device 700 can also be a mobile or stationary server.

[0124] The processor 720 implements the steps of the object interaction method when executing the computer program or instruction.

[0125] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the object interaction method described above are of the same concept. For details not described in detail in the technical scheme of the computing device, please refer to the description of the technical scheme of the object interaction method described above.

[0126] An embodiment of the present specification further provides a computer-readable storage medium storing a computer program or instruction. When the computer program or instruction is executed by a processor, the steps of the object interaction method described above are implemented.

[0127] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of this storage medium and the technical scheme of the object interaction method described above are based on the same concept. For details not described in detail in the technical scheme of the storage medium, please refer to the description of the technical scheme of the object interaction method described above.

[0128] An embodiment of the present specification further provides a computer program product, including a computer program or instructions, which implements the steps of the above-mentioned object interaction method when executed by a processor.

[0129] The above is a schematic solution of a computer program product of this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the object interaction method described above are based on the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the object interaction method described above.

[0130] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0131] The computer program or instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium.

[0132] It should be noted that for the aforementioned method embodiments, for ease of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that this specification is not limited to the order of the actions described, because according to this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this specification.

[0133] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0134] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of this specification, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An object interaction method, characterized in that: include: Determine the interaction instructions submitted by the target object to the auxiliary object in the game scene; determining auxiliary object information, target object information, and scene information of the game scene in response to the interaction instruction, and constructing interaction information based on the auxiliary object information, the target object information, and the scene information; The large language model is used to generate interaction description information corresponding to the interaction information, and the auxiliary object is driven to interact with the target object based on the interaction description information.

2. The object interaction method according to claim 1, characterized in that: In a case where the interaction instruction is a dialogue interaction instruction, determining the auxiliary object information, the target object information, and the scene information of the game scene in response to the interaction instruction includes: collecting auxiliary object attribute information, object memory information, and dialogue style information of the auxiliary object in response to the dialogue interaction instruction, and using the auxiliary object attribute information, the object memory information, and the dialogue style information as the auxiliary object information; In response to the dialogue interaction instruction, target object dialogue information corresponding to the target object is collected, as well as scene information of the game scene, and the target object dialogue information is used as the target object information.

3. The object interaction method according to claim 2, characterized in that: The driving the auxiliary object to interact with the target object based on the interaction description information includes: Constructing a dialogue content that matches the dialogue style information based on the interaction description information, and driving the auxiliary object to interact with the target object based on the dialogue content.

4. The object interaction method according to claim 1, wherein: In a case where the interaction instruction is a behavior interaction instruction, determining the auxiliary object information, the target object information, and the scene information of the game scene in response to the interaction instruction includes: determining a target object behavior and a target object dialogue of the target object, an auxiliary object behavior and an auxiliary object dialogue of the auxiliary object, and a target context of the game scene in response to the behavior interaction instruction; The target object behavior and the target object dialogue are converted into the target object information, the auxiliary object behavior and the auxiliary object dialogue are converted into the auxiliary object information, and the target situation is converted into the scene information.

5. The object interaction method according to claim 4, characterized in that: The driving the auxiliary object to interact with the target object based on the interaction description information includes: Determine interaction behavior parameters based on the interaction description information, and drive the auxiliary object and the target object to complete behavior interaction based on the interaction behavior parameters.

6. The object interaction method according to claim 1, characterized in that: The training of the large language model includes: Determining a pre-trained large language model, and setting the auxiliary object for the game scene; Distributed training is performed on the branch parameters of the pre-trained large language model based on the configuration information of the auxiliary object to obtain a large language model that meets the training stop condition.

7. The object interaction method according to claim 1, characterized in that: The generating of interaction description information corresponding to the interaction information by using a large language model includes: Determining key-value intermediate information associated with the large language model; The large language model is used to predict the interaction information based on the key-value intermediate information to obtain the interaction description information corresponding to the interaction information.

8. The object interaction method according to claim 7, characterized in that: The determining of the key-value intermediate information associated with the large language model includes: Determining initial key-value intermediate information associated with the large language model; The initial key-value intermediate information is divided according to a preset division rule to obtain the key-value intermediate information.

9. An object interaction device, characterized in that: include: A determination module is configured to determine an interaction instruction submitted by a target object to an auxiliary object in a game scene; a construction module configured to determine auxiliary object information, target object information, and scene information of the game scene in response to the interaction instruction, and construct interaction information based on the auxiliary object information, the target object information, and the scene information; A generation module is configured to generate interaction description information corresponding to the interaction information using a large language model, and drive the auxiliary object to interact with the target object based on the interaction description information.

10. A computing device comprising a memory, a processor, and a computer program or instruction stored in the memory and executable on the processor, wherein: When the processor executes the computer program or instructions, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium storing a computer program or instruction, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images