import gradio as gr from gradio_client import Client from huggingface_hub import InferenceClient import random #ss_client = Client("https://omnibus-html-image-current-tab.hf.space/") models=[ "google/gemma-7b", "google/gemma-7b-it", "google/gemma-2b", "google/gemma-2b-it" "meta-llama/Llama-2-7b-chat-hf", "codellama/CodeLlama-70b-Instruct-hf", "openchat/openchat-3.5-0106", "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO", "mistralai/Mixtral-8x7B-Instruct-v0.1", "mistralai/Mixtral-8x7B-Instruct-v0.2" ] '''clients=[ InferenceClient(models[0]), InferenceClient(models[1]), InferenceClient(models[2]), InferenceClient(models[3]), ]''' client_z=[] def load_models(inp): print(type(inp)) client_z.clear() if inp[0]: client_z[0]=InferenceClient(models[inp[0]]) if inp[1]: client_z[1]=InferenceClient(models[inp[1]]) if inp[2]: client_z[2]=InferenceClient(models[inp[2]]) if inp[3]: client_z[3]=InferenceClient(models[inp[3]]) return gr.update(label=models[inp[0]]) def format_prompt(message, history): prompt = "" if history: #userHow does the brain work?model for user_prompt, bot_response in history: prompt += f"{user_prompt}\n" print(prompt) prompt += f"{bot_response}\n" print(prompt) prompt += f"user{message}model" print(prompt) return prompt def chat_inf(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p): #token max=8192 client=clients[int(client_choice)-1] if not history: history = [] hist_len=0 if history: hist_len=len(history) print(hist_len) in_len=len(system_prompt+prompt)+hist_len print("\n#########"+in_len) if (in_len+tokens) > 8000: yield [(prompt,"Wait. I need to compress our Chat history...")] history=compress_history(history,client_choice,seed,temp,tokens,top_p,rep_p) yield [(prompt,"History has been compressed, processing request...")] generate_kwargs = dict( temperature=temp, max_new_tokens=tokens, top_p=top_p, repetition_penalty=rep_p, do_sample=True, seed=seed, ) #formatted_prompt=prompt 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 = "" for response in stream: output += response.token.text yield [(prompt,output)] history.append((prompt,output)) yield history def clear_fn(): return None,None,None rand_val=random.randint(1,1111111111111111) def check_rand(inp,val): if inp==True: return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1,1111111111111111)) else: return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val)) with gr.Blocks() as app: gr.HTML("""

Google Gemma Models


running on Huggingface Inference Client


EXPERIMENTAL""") with gr.Row(): chat_a = gr.Chatbot(height=500) chat_b = gr.Chatbot(height=500) with gr.Row(): chat_c = gr.Chatbot(height=500) chat_d = gr.Chatbot(height=500) with gr.Group(): with gr.Row(): with gr.Column(scale=3): inp = gr.Textbox(label="Prompt") sys_inp = gr.Textbox(label="System Prompt (optional)") with gr.Row(): with gr.Column(scale=2): btn = gr.Button("Chat") with gr.Column(scale=1): with gr.Group(): stop_btn=gr.Button("Stop") clear_btn=gr.Button("Clear") client_choice=gr.Dropdown(label="Models",type='index',choices=[c for c in models],value=models[0],multiselect=True,interactive=True) with gr.Column(scale=1): with gr.Group(): rand = gr.Checkbox(label="Random Seed", value=True) seed=gr.Slider(label="Seed", minimum=1, maximum=1111111111111111,step=1, value=rand_val) tokens = gr.Slider(label="Max new tokens",value=3840,minimum=0,maximum=8000,step=64,interactive=True, visible=True,info="The maximum number of tokens") temp=gr.Slider(label="Temperature",step=0.01, minimum=0.01, maximum=1.0, value=0.9) top_p=gr.Slider(label="Top-P",step=0.01, minimum=0.01, maximum=1.0, value=0.9) rep_p=gr.Slider(label="Repetition Penalty",step=0.1, minimum=0.1, maximum=2.0, value=1.0) with gr.Accordion(label="Screenshot",open=False): with gr.Row(): with gr.Column(scale=3): im_btn=gr.Button("Screenshot") img=gr.Image(type='filepath') with gr.Column(scale=1): with gr.Row(): im_height=gr.Number(label="Height",value=5000) im_width=gr.Number(label="Width",value=500) wait_time=gr.Number(label="Wait Time",value=3000) theme=gr.Radio(label="Theme", choices=["light","dark"],value="light") chatblock=gr.Dropdown(label="Chatblocks",info="Choose specific blocks of chat",choices=[c for c in range(1,40)],multiselect=True) client_choice.change(load_models,client_choice,chat_a) #im_go=im_btn.click(get_screenshot,[chat_b,im_height,im_width,chatblock,theme,wait_time],img) #chat_sub=inp.submit(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,client_choice,seed,temp,tokens,top_p,rep_p],chat_b) #go=btn.click(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,client_choice,seed,temp,tokens,top_p,rep_p],chat_b) #stop_btn.click(None,None,None,cancels=[go,im_go,chat_sub]) #clear_btn.click(clear_fn,None,[inp,sys_inp,chat_b]) app.queue(default_concurrency_limit=10).launch()