Merge pull request #104 from yukiarimo/dev

Release V6.5.4

lib/audio.py

@@ -46,7 +46,7 @@ def create_speaker_embedding(waveform):
             speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy()
         return speaker_embeddings
 
-    audio_array, sampling_rate = librosa.load("/Users/yuki/Documents/Github/yuna-ai/static/audio/" + config["server"]["yuna_audio_name"], sr=16000)
+    audio_array, sampling_rate = librosa.load("/Users/yuki/Documents/Github/yuna-ai/static/audio/" + config["server"]["yuna_reference_audio"], sr=16000)
 
     # Create a dictionary to mimic the dataset structure
     custom_audio = {
@@ -157,7 +157,7 @@ def speak_text(text, reference_audio=config['server']['yuna_reference_audio'], o
         audio = AudioSegment.from_file("/Users/yuki/Documents/Github/yuna-ai/static/audio/audio.aiff")
         audio.export("/Users/yuki/Documents/Github/yuna-ai/static/audio/audio.mp3", format='mp3')
     elif mode == "siri-pv":
-        command = f'say -v {config["server"]["yuna_audio_name"]} -o /Users/yuki/Documents/Github/yuna-ai/static/audio/audio.aiff {repr(text)}'
+        command = f'say -v {config["server"]["yuna_reference_audio"]} -o /Users/yuki/Documents/Github/yuna-ai/static/audio/audio.aiff {repr(text)}'
         print(command)
         exit_status = os.system(command)
 

lib/generate.py

@@ -21,7 +21,7 @@ def get_config(config=None):
     from langchain_community.document_loaders import TextLoader
     from langchain.text_splitter import RecursiveCharacterTextSplitter
     from langchain_community.vectorstores import Chroma
-    from langchain_community.embeddings import GPT4AllEmbeddings
+    from langchain_huggingface import HuggingFaceEmbeddings
     from langchain.chains import RetrievalQA
     from langchain_community.llms import LlamaCpp
 
@@ -31,12 +31,15 @@ def __init__(self, config):
         self.model = Llama(
             model_path="lib/models/yuna/" + config["server"]["yuna_default_model"],
             n_ctx=config["ai"]["context_length"],
+            last_n_tokens_size=config["ai"]["last_n_tokens_size"],
             seed=config["ai"]["seed"],
             n_batch=config["ai"]["batch_size"],
             n_gpu_layers=config["ai"]["gpu_layers"],
             n_threads=config["ai"]["threads"],
+            use_mmap=config["ai"]["use_mmap"],
             use_mlock=config["ai"]["use_mlock"],
             flash_attn=config["ai"]["flash_attn"],
+            offload_kqv=config["ai"]["offload_kqv"],
             verbose=False
         ) if config["server"]["yuna_text_mode"] == "native" else ""
 
@@ -173,7 +176,7 @@ def generate(self, chat_id, speech=False, text="", template=None, chat_history_m
             return ''.join(response) if isinstance(response, (list, type((lambda: (yield))()))) else response
         return response
 
-    def processTextFile(self, text_file, question, temperature=0.8, max_new_tokens=128, context_window=256):
+    def processTextFile(self, text_file, question, temperature):
         # Load text file data
         loader = TextLoader(text_file)
         data = loader.load()
@@ -182,16 +185,14 @@ def processTextFile(self, text_file, question, temperature=0.8, max_new_tokens=1
         text_splitter = RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=0)
         docs = text_splitter.split_documents(data)
 
-        # Generate embeddings locally using GPT4All
-        gpt4all_embeddings = GPT4AllEmbeddings()
-        vectorstore = Chroma.from_documents(documents=docs, embedding=gpt4all_embeddings)
+        # Generate embeddings using HuggingFaceEmbeddings
+        huggingface_embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
+        vectorstore = Chroma.from_documents(documents=docs, embedding=huggingface_embeddings)
 
         # Load GPT4All model for inference  
         llm = LlamaCpp(
-            model_path="lib/models/yuna/" + self.config["server"]["yuna_default_model"],
+            model_path="lib/models/yuna/yuna-ai-v3-q5_k_m.gguf",
             temperature=temperature,
-            max_new_tokens=max_new_tokens,
-            context_window=context_window,
             verbose=False,
         )
 
@@ -200,7 +201,7 @@ def processTextFile(self, text_file, question, temperature=0.8, max_new_tokens=1
 
         # Ask a question 
         result = qa.invoke(question)
-        return result
+        return result['result']
 
     def get_history_text(self, chat_history, text, useHistory, yunaConfig):
         history = ''

lib/router.py

@@ -215,7 +215,7 @@ def handle_textfile_request(chat_generator, self):
     query = request.form['query']
     text_file.save('static/text/content.txt')
 
-    result = chat_generator.processTextFile('static/text/content.txt', query)
+    result = chat_generator.processTextFile('static/text/content.txt', query, 0.6)
 
     return jsonify({'response': result})
 

lib/vision.py

@@ -10,13 +10,16 @@
     model_path="lib/models/agi/miru/" + config["server"]["miru_default_model"],
     chat_handler=MoondreamChatHandler(clip_model_path="lib/models/agi/miru/" + config["server"]["eyes_default_model"]),
     n_ctx=4096,
+    last_n_tokens_size=config["ai"]["last_n_tokens_size"],
     seed=config["ai"]["seed"],
     n_batch=config["ai"]["batch_size"],
     n_gpu_layers=config["ai"]["gpu_layers"],
     n_threads=config["ai"]["threads"],
+    use_mmap=config["ai"]["use_mmap"],
     use_mlock=config["ai"]["use_mlock"],
     flash_attn=config["ai"]["flash_attn"],
-    verbose=False,
+    offload_kqv=config["ai"]["offload_kqv"],
+    verbose=False
 ) if config["ai"]["miru"] == True else ""
 
 if config["ai"]["art"] == True:
@@ -50,14 +53,14 @@ def capture_image(image_path=None, prompt=None, use_cpu=False, speech=False):
 
     result = llm.create_chat_completion(
     messages = [
-        {"role": "system", "content": "You are an assistant who perfectly describes images and answers questions about them."},
-        {
-            "role": "user",
-            "content": [
-                {"type" : "text", "text": prompt},
-                {"type": "image_url", "image_url": {"url": "file://" + abs_image_path } }
-            ]
-        }
+            {"role": "system", "content": "You are an assistant who perfectly describes images and answers questions about them."},
+            {
+                "role": "user",
+                "content": [
+                    {"type" : "text", "text": prompt},
+                    {"type": "image_url", "image_url": {"url": "file://" + abs_image_path } }
+                ]
+            }
         ]
     )
 

static/config.json

@@ -4,17 +4,17 @@
             "Yuki",
             "Yuna"
         ],
-        "hinmitsu": false,
+        "himitsu": false,
         "agi": false,
         "emotions": false,
         "art": false,
         "miru": false,
         "search": false,
-        "max_new_tokens": 512,
-        "context_length": 2048,
+        "max_new_tokens": 256,
+        "context_length": 1536,
         "temperature": 0.7,
         "repetition_penalty": 1.1,
-        "last_n_tokens": 128,
+        "last_n_tokens_size": 128,
         "seed": -1,
         "top_k": 100,
         "top_p": 0.92,
@@ -33,14 +33,14 @@
         "batch_size": 512,
         "threads": 8,
         "gpu_layers": -1,
+        "use_mmap": true,
         "flash_attn": true,
-        "use_mlock": true
+        "use_mlock": true,
+        "offload_kqv": true
     },
     "server": {
         "port": "",
         "url": "",
-        "default_history_file": "history_template:general.json",
-        "images": "images/",
         "yuna_default_model": "yuna-ai-v3-q5_k_m.gguf",
         "miru_default_model": "yuna-ai-miru-v0.gguf",
         "eyes_default_model": "yuna-ai-miru-eye-v0.gguf",
@@ -49,10 +49,19 @@
         "device": "mps",
         "yuna_text_mode": "native",
         "yuna_audio_mode": "siri",
-        "yuna_audio_name": "1.wav",
-        "yuna_reference_audio": "audio.mp3",
+        "yuna_reference_audio": "1.wav",
         "output_audio_format": "audio.aiff"
     },
+    "settings": {
+        "streaming": true,
+        "customConfig": true,
+        "default_history_file": "history_template:general.json",
+        "default_kanojo": "Yuna",
+        "default_prompt_template": "dialog",
+        "background_call": false,
+        "nsfw_filter": false,
+        "dark_mode": true
+    },
     "security": {
         "secret_key": "YourSecretKeyHere123!",
         "encryption_key": "zWZnu-lxHCTgY_EqlH4raJjxNJIgPlvXFbdk45bca_I=",

static/css/index.css

@@ -33,7 +33,7 @@ body {
     height: 100vh !important;
 }
 
-#profile-card {
+.profile-card {
     overflow-y: scroll;
     height: 100vh;
 }
@@ -185,7 +185,6 @@ button {
 .settings-container {
     margin: 50px auto;
     background-color: #1f1f1f;
-    padding: 20px;
     border-radius: 8px;
     margin-top: 50px !important;
 }

static/js/creator.js

@@ -1,28 +1,26 @@
 // Get the necessary elements
-const promptTemplateTextarea = document.querySelector('#article-prompt-template');
-const bodyTextTextarea = document.querySelector('#body-text-article-container');
-const resultTextarea = document.querySelector('#result-create-article');
-const submitButton = document.getElementById('send-create-article');
+const promptTemplateTextareaArticle = document.querySelector('#article-prompt-template');
+const bodyTextTextareaArticle = document.querySelector('#body-text-article-container');
+const resultTextareaArticle = document.querySelector('#result-create-article');
+const submitButtonArticle = document.getElementById('send-create-article');
 
 // Set the default text for the Prompt Template block
 const defaultPromptTemplate = promptTemplateManager.buildPrompt('himitsuAssistant');
-promptTemplateTextarea.value = defaultPromptTemplate.replace('### Instruction:\n', '### Instruction:\n{body_text}');
+promptTemplateTextareaArticle.value = defaultPromptTemplate.replace('### Instruction:\n', '### Instruction:\n{body_text}');
 
 // Function to send the request to the server
-async function sendRequest() {
-    activeElement = document.getElementById('body-text-article-container');
-
-    const bodyText = bodyTextTextarea.value;
-    const promptTemplate = promptTemplateTextarea.value.replace('{body_text}', bodyText);
+async function sendRequestArticle() {
+    const bodyText = bodyTextTextareaArticle.value;
+    const promptTemplate = promptTemplateTextareaArticle.value.replace('{body_text}', bodyText);
 
     // Clear the result textarea before starting
-    resultTextarea.value = '';
+    resultTextareaArticle.value = '';
 
-    messageManagerInstance.sendMessage(promptTemplate, null, imageData = '', url = '/message', naked = false, stream = true, outputElement = resultTextarea);
+    messageManagerInstance.sendMessage(promptTemplate, null, imageData = '', url = '/message', naked = false, stream = true, outputElement = resultTextareaArticle);
 }
 
 // Add an event listener to the submit button
-submitButton.addEventListener('click', sendRequest);
+submitButtonArticle.addEventListener('click', sendRequestArticle);
 
 // Get the necessary elements for the Presentation tab
 const presentationPromptTemplateTextarea = document.getElementById('presentation-prompt-template');
@@ -38,8 +36,6 @@ presentationPromptTemplateTextarea.value = defaultPresentationPromptTemplate;
 
 // Function to send the request to the server for the Presentation tab
 async function sendPresentationRequest() {
-    activeElement = document.getElementById('user-input-presentation-container');
-
     const userInput = presentationUserInputTextarea.value;
     const promptTemplate = presentationPromptTemplateTextarea.value.replace('{user_input}', userInput);
 
@@ -56,4 +52,16 @@ function copyToDraft() {
 
 // Add event listeners to the buttons
 generateButton.addEventListener('click', sendPresentationRequest);
-copyToDraftButton.addEventListener('click', copyToDraft);
+copyToDraftButton.addEventListener('click', copyToDraft);
+
+// Function for naked mode
+const nakedWorkArea = document.querySelector('#naked-work-area');
+const submitButtonNaked = document.getElementById('send-create-naked');
+
+// Function to send the request to the server
+async function sendRequestNaked() {
+    messageManagerInstance.sendMessage(nakedWorkArea.value, null, imageData = '', url = '/message', naked = false, stream = true, outputElement = nakedWorkArea);
+}
+
+// Add an event listener to the submit button
+submitButtonNaked.addEventListener('click', sendRequestNaked);