app.py

import gradio as gr
from huggingface_hub import InferenceClient


client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
#client = InferenceClient("Qwen/Qwen2.5-7B-Instruct")


def generate_text(messages):
    print("generate_text")
    print(messages)
    generated = ""
    for token in client.chat_completion(messages, max_tokens=100,stream=True):
        content = (token.choices[0].delta.content)
        generated += content
        #print(content)
        #print(''.join(list(content)))
        yield generated
        #print(token.choices[0].delta)
    
    #return generated+"\n" #no stram version

def call_generate_text(message, history):
    #if len(message) == 0:
    #    messages.append({"role": "system", "content": "you response around 10 words"})
   
    print(message)
    print(history)

    user_message = [{"role":"user","content":message}]
    messages = history + user_message
    try:
        
        assistant_message={"role":"assistant","content":""}
        text_generator = generate_text(messages)

        for text_chunk in text_generator:
            print(f"chunk={text_chunk}")
            assistant_message["content"] = text_chunk
            updated_history = messages + [assistant_message]
            yield "", updated_history

    except RuntimeError  as e:
        print(f"An unexpected error occurred: {e}")
        yield  "",history

head = '''
<script src="https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.webgpu.min.js" ></script>
<script type="module">
        import { env } from "http://localhost:8000/js-esm/matcha_tts_onnx_en.js";
        env.matcha_tts_model_path = "/file=models/ljspeech_sim.onnx"
</script>
'''

with gr.Blocks(title="LLM with TTS",head=head) as demo:
    gr.Markdown("## A LLM is unstable:The inference client used in this demo exhibits inconsistent performance. While it can provide responses in milliseconds, it sometimes becomes unresponsive and times out.")
    gr.Markdown("### TTS talke a long loading time:Please be patient, the first response may have a delay of up to over 40 seconds while loading.")
    gr.Markdown("**Mistral-7B-Instruct-v0.3/LJSpeech**.LLM and TTS models will change without notice.")
    
    js = """
    function(chatbot){
        window.update_chatbot(chatbot)
        //auto scroll
        var chatElement = document.getElementById('gr-chatbot');
        chatElement.scrollTop = chatElement.scrollHeight;
        var logElement = chatElement.querySelector('div[role="log"]');
        logElement.scrollTop = logElement.scrollHeight;
    }
    """
    chatbot = gr.Chatbot(type="messages",elem_id="gr-chatbot")
    chatbot.change(None,[chatbot],[],js=js)
    msg = gr.Textbox()
    with gr.Row():
        clear = gr.ClearButton([msg, chatbot])
        submit = gr.Button("Submit",variant="primary").click(call_generate_text, inputs=[msg, chatbot], outputs=[msg,chatbot])

    gr.HTML("""
    <br>
    <div id="footer">
    <b>Spaces</b><br>
     <a href="https://huggingface.co/spaces/Akjava/matcha-tts_vctk-onnx" style="font-size: 9px" target="link">Match-TTS VCTK-ONNX</a> | 
     <a href="https://huggingface.co/spaces/Akjava/matcha-tts-onnx-benchmarks" style="font-size: 9px" target="link">Match-TTS ONNX-Benchmark</a> | 
     <a href="https://huggingface.co/spaces/Akjava/AIChat-matcha-tts-onnx-en" style="font-size: 9px" target="link">AIChat-Matcha-TTS ONNX English</a> | 
     
      <br><br>
    <b>Credits</b><br>
    <a href="https://github.com/akjava/Matcha-TTS-Japanese" style="font-size: 9px" target="link">Matcha-TTS-Japanese</a> | 
    <a href = "http://www.udialogue.org/download/cstr-vctk-corpus.html" style="font-size: 9px"  target="link">CSTR VCTK Corpus</a> |
    <a href = "https://github.com/cmusphinx/cmudict" style="font-size: 9px"  target="link">CMUDict</a> |
    <a href = "https://huggingface.co/docs/transformers.js/index" style="font-size: 9px"  target="link">Transformer.js</a> |
    <a href = "https://huggingface.co/cisco-ai/mini-bart-g2p" style="font-size: 9px"  target="link">mini-bart-g2p</a> |
    <a href = "https://onnxruntime.ai/docs/get-started/with-javascript/web.html" style="font-size: 9px"  target="link">ONNXRuntime-Web</a> |
    <a href = "https://github.com/akjava/English-To-IPA-Collections" style="font-size: 9px"  target="link">English-To-IPA-Collections</a> |
    <a href ="https://huggingface.co/papers/2309.03199" style="font-size: 9px"  target="link">Matcha-TTS Paper</a>
    </div>
    """)
    
    msg.submit(call_generate_text, [msg, chatbot], [msg, chatbot])

import os
remote_dir ="/home/user/app/"
local_dir = "C:\\Users\\owner\\Documents\\pythons\\huggingface\\mistral-7b-v0.3-matcha-tts-en"  #sorry this is my develop env
#print(os.path.join(dir,"js-esm","matcha_tts_onnx_en.js"))
# set not dir but file
#demo.launch(allowed_paths=[os.path.join(remote_dir,"models","ljspeech_sim.onnx")])
demo.launch(allowed_paths=[os.path.join(local_dir,"models","ljspeech_sim.onnx")])