Suppose to implement more TTS chains.

A lot of api changed in Unity.Sentis 1.4-pre.

Issue Description

When retrieval is enabled, bot will appear with duplicate answers.

Expected Behavior

Bot should not repeat its response in the same session.

Proposed Solution

Modify filter，or add a preprocessor before it. So we should only allocate tensors among answers that have not been answered.

Pipeline's encoder need load model and tokenizer, currently can only read file from UserData/models.

Suppose to implement STT chains such as whisper.

Langchain is a powerful tool to create agent-based workflow. It will be interesting to implement it in Unity for game AI usage.🥰
I will first create a tool use system for quick verification.

Consider to use finite state machine.

以下是代码
public class RobotController : MonoBehaviour
{
public TMP_InputField inputField;
public TMP_Text TMP_Text;
public string prompt = @"How many planets are there?";
public AudioSource speechAudioSource;

private ChatHistory history;
private ChatMemory memory;
private readonly string template = "{history}\n{prompt}";
public ILargeLanguageModel ollamaChat;
// Start is called before the first frame update
void Start()
{
    ollamaChat = new OllamaChat() { Model = "BobBot" };
    history = new ChatHistory();
    memory = new ChatBufferMemory() { ChatHistory = history};
    inputField.onValueChanged.AddListener((str) => {
        prompt = str;
    });
}

public void OnClick() {
    RunChainlocal();
}

public async UniTask RunChainlocal()
{
    var chain = Chain.Set(prompt, "prompt")
        | Chain.LoadMemory(memory,"history")
        | Chain.Template(template)
        | Chain.LLM(ollamaChat)
        | Chain.UpdateHistory(history,requestKey:"prompt");

    string result = await chain.Run<string>("text");
    TMP_Text.text = result.Replace("\n", string.Empty).Trim();
    //    //AzureTTSStream.Instance.StartTTS(result, speechAudioSource);
    //    //AzureTTSNormal.Instance.StartTTS(result, speechAudioSource);
}

}
以下是log
[chain/start] [1:chain:StackChain > 9:chain:UpdateHistoryChain] Entering chain run with input:
{"prompt":"我叫什么名字","history":"User: 你好，我叫姚大哥，请你记住我\r\nBot: 好的，很高兴认识你！请告诉我你的名字是什么？\r\nUser: 我的名字是姚大哥\r\nBot: 好的，小姚大哥。很高兴与你交谈！有事可以随时找我聊天哦。\r\n","text":"很抱歉，我不知道你的名字。如果你告诉我，你就能听到我的回答了！"}

[chain/end] [1:chain:StackChain > 9:chain:UpdateHistoryChain] [1.0067ms] Exiting chain run with output:
{"prompt":"我叫什么名字","history":"User: 你好，我叫姚大哥，请你记住我\r\nBot: 好的，很高兴认识你！请告诉我你的名字是什么？\r\nUser: 我的名字是姚大哥\r\nBot: 好的，小姚大哥。很高兴与你交谈！有事可以随时找我聊天哦。\r\n","text":"很抱歉，我不知道你的名字。如果你告诉我，你就能听到我的回答了！"}

[chain/end] [1:chain:StackChain] [786.7143ms] Exiting chain run with output:
{"prompt":"我叫什么名字","history":"User: 你好，我叫姚大哥，请你记住我\r\nBot: 好的，很高兴认识你！请告诉我你的名字是什么？\r\nUser: 我的名字是姚大哥\r\nBot: 好的，小姚大哥。很高兴与你交谈！有事可以随时找我聊天哦。\r\n","text":"很抱歉，我不知道你的名字。如果你告诉我，你就能听到我的回答了！"}