LLMCaller.cs

Go to the documentation of this file.

 
using System;
using System.Collections.Generic;
using System.Threading.Tasks;
using UnityEngine;
using UnityEngine.Networking;
 
namespace LLMUnity
{
    [DefaultExecutionOrder(-2)]
    public class LLMCaller : MonoBehaviour
    {
        [HideInInspector] public bool advancedOptions = false;
        [LocalRemote] public bool remote = false;
        [Local, SerializeField] protected LLM _llm;
        public LLM llm
        {
            get => _llm;//whatever
            set => SetLLM(value);
        }
 
        [Remote] public string APIKey;
 
        [Remote] public string host = "localhost";
        [Remote] public int port = 13333;
        [Remote] public int numRetries = 10;
 
        protected LLM _prellm;
        protected List<(string, string)> requestHeaders;
        protected List<UnityWebRequest> WIPRequests = new List<UnityWebRequest>();
 
        public virtual void Awake()
        {
            // Start the LLM server in a cross-platform way
            if (!enabled) return;
 
            requestHeaders = new List<(string, string)> { ("Content-Type", "application/json") };
            if (!remote)
            {
                AssignLLM();
                if (llm == null)
                {
                    string error = $"No LLM assigned or detected for LLMCharacter {name}!";
                    LLMUnitySetup.LogError(error);
                    throw new Exception(error);
                }
            }
            else
            {
                if (!String.IsNullOrEmpty(APIKey)) requestHeaders.Add(("Authorization", "Bearer " + APIKey));
            }
        }
 
        protected virtual void SetLLM(LLM llmSet)
        {
            if (llmSet != null && !IsValidLLM(llmSet))
            {
                LLMUnitySetup.LogError(NotValidLLMError());
                llmSet = null;
            }
            _llm = llmSet;
            _prellm = _llm;
        }
 
        public virtual bool IsValidLLM(LLM llmSet)
        {
            return true;
        }
 
        public virtual bool IsAutoAssignableLLM(LLM llmSet)
        {
            return true;
        }
 
        protected virtual string NotValidLLMError()
        {
            return $"Can't set LLM {llm.name} to {name}";
        }
 
        protected virtual void OnValidate()
        {
            if (_llm != _prellm) SetLLM(_llm);
            AssignLLM();
        }
 
        protected virtual void Reset()
        {
            AssignLLM();
        }
 
        protected virtual void AssignLLM()
        {
            if (remote || llm != null) return;
 
            List<LLM> validLLMs = new List<LLM>();
            foreach (LLM foundllm in FindObjectsOfType<LLM>())
            {
                if (IsValidLLM(foundllm) && IsAutoAssignableLLM(foundllm)) validLLMs.Add(foundllm);
            }
            if (validLLMs.Count == 0) return;
 
            llm = SortLLMsByBestMatching(validLLMs.ToArray())[0];
            string msg = $"Assigning LLM {llm.name} to {GetType()} {name}";
            if (llm.gameObject.scene != gameObject.scene) msg += $" from scene {llm.gameObject.scene}";
            LLMUnitySetup.Log(msg);
        }
 
        protected virtual LLM[] SortLLMsByBestMatching(LLM[] arrayIn)
        {
            LLM[] array = (LLM[])arrayIn.Clone();
            for (int i = 0; i < array.Length - 1; i++)
            {
                bool swapped = false;
                for (int j = 0; j < array.Length - i - 1; j++)
                {
                    bool sameScene = array[j].gameObject.scene == array[j + 1].gameObject.scene;
                    bool swap = (
                        (!sameScene && array[j + 1].gameObject.scene == gameObject.scene) ||
                        (sameScene && array[j].transform.GetSiblingIndex() > array[j + 1].transform.GetSiblingIndex())
                    );
                    if (swap)
                    {
                        LLM temp = array[j];
                        array[j] = array[j + 1];
                        array[j + 1] = temp;
                        swapped = true;
                    }
                }
                if (!swapped) break;
            }
            return array;
        }
 
        protected virtual List<int> TokenizeContent(TokenizeResult result)
        {
            // get the tokens from a tokenize result received from the endpoint
            return result.tokens;
        }
 
        protected virtual string DetokenizeContent(TokenizeRequest result)
        {
            // get content from a chat result received from the endpoint
            return result.content;
        }
 
        protected virtual List<float> EmbeddingsContent(EmbeddingsResult result)
        {
            // get content from a chat result received from the endpoint
            return result.embedding;
        }
 
        protected virtual Ret ConvertContent<Res, Ret>(string response, ContentCallback<Res, Ret> getContent = null)
        {
            // template function to convert the json received and get the content
            if (response == null) return default;
            response = response.Trim();
            if (response.StartsWith("data: "))
            {
                string responseArray = "";
                foreach (string responsePart in response.Replace("\n\n", "").Split("data: "))
                {
                    if (responsePart == "") continue;
                    if (responseArray != "") responseArray += ",\n";
                    responseArray += responsePart;
                }
                response = $"{{\"data\": [{responseArray}]}}";
            }
            return getContent(JsonUtility.FromJson<Res>(response));
        }
 
        protected virtual void CancelRequestsLocal() {}
 
        protected virtual void CancelRequestsRemote()
        {
            foreach (UnityWebRequest request in WIPRequests)
            {
                request.Abort();
            }
            WIPRequests.Clear();
        }
 
        // <summary>
        public virtual void CancelRequests()
        {
            if (remote) CancelRequestsRemote();
            else CancelRequestsLocal();
        }
 
        protected virtual async Task<Ret> PostRequestLocal<Res, Ret>(string json, string endpoint, ContentCallback<Res, Ret> getContent, Callback<Ret> callback = null)
        {
            // send a post request to the server and call the relevant callbacks to convert the received content and handle it
            // this function has streaming functionality i.e. handles the answer while it is being received
            while (!llm.failed && !llm.started) await Task.Yield();
            string callResult = null;
            switch (endpoint)
            {
                case "tokenize":
                    callResult = await llm.Tokenize(json);
                    break;
                case "detokenize":
                    callResult = await llm.Detokenize(json);
                    break;
                case "embeddings":
                    callResult = await llm.Embeddings(json);
                    break;
                case "slots":
                    callResult = await llm.Slot(json);
                    break;
                default:
                    LLMUnitySetup.LogError($"Unknown endpoint {endpoint}");
                    break;
            }
 
            Ret result = ConvertContent(callResult, getContent);
            callback?.Invoke(result);
            return result;
        }
 
        protected virtual async Task<Ret> PostRequestRemote<Res, Ret>(string json, string endpoint, ContentCallback<Res, Ret> getContent, Callback<Ret> callback = null)
        {
            // send a post request to the server and call the relevant callbacks to convert the received content and handle it
            // this function has streaming functionality i.e. handles the answer while it is being received
            if (endpoint == "slots")
            {
                LLMUnitySetup.LogError("Saving and loading is not currently supported in remote setting");
                return default;
            }
 
            Ret result = default;
            byte[] jsonToSend = new System.Text.UTF8Encoding().GetBytes(json);
            UnityWebRequest request = null;
            string error = null;
            int tryNr = numRetries;
 
            while (tryNr != 0)
            {
                using (request = UnityWebRequest.Put($"{host}:{port}/{endpoint}", jsonToSend))
                {
                    WIPRequests.Add(request);
 
                    request.method = "POST";
                    if (requestHeaders != null)
                    {
                        for (int i = 0; i < requestHeaders.Count; i++)
                            request.SetRequestHeader(requestHeaders[i].Item1, requestHeaders[i].Item2);
                    }
 
                    // Start the request asynchronously
                    var asyncOperation = request.SendWebRequest();
                    float lastProgress = 0f;
                    // Continue updating progress until the request is completed
                    while (!asyncOperation.isDone)
                    {
                        float currentProgress = request.downloadProgress;
                        // Check if progress has changed
                        if (currentProgress != lastProgress && callback != null)
                        {
                            callback?.Invoke(ConvertContent(request.downloadHandler.text, getContent));
                            lastProgress = currentProgress;
                        }
                        // Wait for the next frame
                        await Task.Yield();
                    }
                    WIPRequests.Remove(request);
                    if (request.result == UnityWebRequest.Result.Success)
                    {
                        result = ConvertContent(request.downloadHandler.text, getContent);
                        error = null;
                        break;
                    }
                    else
                    {
                        result = default;
                        error = request.error;
                        if (request.responseCode == (int)System.Net.HttpStatusCode.Unauthorized) break;
                    }
                }
                tryNr--;
                if (tryNr > 0) await Task.Delay(200 * (numRetries - tryNr));
            }
 
            if (error != null) LLMUnitySetup.LogError(error);
            callback?.Invoke(result);
            return result;
        }
 
        protected virtual async Task<Ret> PostRequest<Res, Ret>(string json, string endpoint, ContentCallback<Res, Ret> getContent, Callback<Ret> callback = null)
        {
            if (remote) return await PostRequestRemote(json, endpoint, getContent, callback);
            return await PostRequestLocal(json, endpoint, getContent, callback);
        }
 
        public virtual async Task<List<int>> Tokenize(string query, Callback<List<int>> callback = null)
        {
            // handle the tokenization of a message by the user
            TokenizeRequest tokenizeRequest = new TokenizeRequest();
            tokenizeRequest.content = query;
            string json = JsonUtility.ToJson(tokenizeRequest);
            return await PostRequest<TokenizeResult, List<int>>(json, "tokenize", TokenizeContent, callback);
        }
 
        public virtual async Task<string> Detokenize(List<int> tokens, Callback<string> callback = null)
        {
            // handle the detokenization of a message by the user
            TokenizeResult tokenizeRequest = new TokenizeResult();
            tokenizeRequest.tokens = tokens;
            string json = JsonUtility.ToJson(tokenizeRequest);
            return await PostRequest<TokenizeRequest, string>(json, "detokenize", DetokenizeContent, callback);
        }
 
        public virtual async Task<List<float>> Embeddings(string query, Callback<List<float>> callback = null)
        {
            // handle the tokenization of a message by the user
            TokenizeRequest tokenizeRequest = new TokenizeRequest();
            tokenizeRequest.content = query;
            string json = JsonUtility.ToJson(tokenizeRequest);
            return await PostRequest<EmbeddingsResult, List<float>>(json, "embeddings", EmbeddingsContent, callback);
        }
    }
}