from huggingface_hub import InferenceClient import gradio as gr import random import prompts from pypipertts import PyPiper pp=PyPiper() pp.load_mod(instr="en_US-joe-medium") client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1") def format_prompt(message, history): prompt = "" for user_prompt, bot_response in history: print (bot_response) prompt += f"[INST] {user_prompt} [/INST]" prompt += f" {bot_response} " prompt += f"[INST] {message} [/INST]" return prompt def generate(prompt,history,max_new_tokens,seed): #seed = random.randint(1,9999999999999) print(seed) system_prompt = prompts.AI_REPORT_WRITER generate_kwargs = dict( temperature=0.9, max_new_tokens=max_new_tokens, top_p=0.95, repetition_penalty=1.0, do_sample=True, seed=seed, ) formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history) stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False) output = "" buf = "" for response in stream: if response.token.text == "\n": yield from pp.stream_tts(buf) buf="" buf += response.token.text output += response.token.text print(output) #yield output with gr.Blocks() as iface: gr.HTML("""""") aud=gr.Audio(streaming=True,autoplay=True) #chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=True, likeable=True, layout="panel"), chatbot=gr.Chatbot() msg = gr.Textbox() with gr.Row(): submit_b = gr.Button() stop_b = gr.Button("Stop") clear = gr.ClearButton([msg, chatbot]) sumbox=gr.Textbox("Summary", max_lines=100) with gr.Column(): sum_out_box=gr.JSON(label="Summaries") hist_out_box=gr.JSON(label="History") #sub_b = submit_b.click(generate, [msg,chatbot],[msg,chatbot,sumbox,sum_out_box,hist_out_box]) #sub_e = msg.submit(generate, [msg, chatbot], [msg, chatbot,sumbox,sum_out_box,hist_out_box]) sub_b = submit_b.click(generate, [msg,chatbot],aud) sub_e = msg.submit(generate, [msg, chatbot], aud) stop_b.click(None,None,None, cancels=[sub_b,sub_e]) iface.queue(default_concurrency_limit=10).launch()