File size: 6,242 Bytes
04c25c5
c745c39
04c25c5
9b9128d
c745c39
 
1d73b44
 
 
 
 
 
 
 
 
 
 
 
c745c39
5d5363d
c745c39
e9b47ff
d786de6
e9b47ff
8686afb
e9b47ff
5d5363d
 
 
 
8686afb
e9b47ff
1d73b44
5d5363d
369dc1f
5d5363d
2a37133
04c25c5
 
 
5d5363d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
04c25c5
c745c39
 
 
 
 
 
 
 
 
 
d7942b7
5d5363d
61da65e
5d5363d
61da65e
 
 
 
 
 
04c25c5
5d5363d
34a59de
0d86f5d
04c25c5
 
 
 
 
61da65e
 
 
 
 
 
 
0d86f5d
04c25c5
 
0d86f5d
61da65e
5d5363d
61da65e
 
 
5d5363d
c745c39
 
 
 
 
 
 
 
 
 
 
 
5d5363d
9b2cb9f
5d5363d
 
64dcdf0
5d5363d
d0d4c9d
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
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"
]
clients=[
InferenceClient(models[0]),
InferenceClient(models[1]),
InferenceClient(models[2]),
InferenceClient(models[3]),
]

VERBOSE=True

def format_prompt(message, history):
    prompt = ""
    if history:
        #<start_of_turn>userHow does the brain work?<end_of_turn><start_of_turn>model
        for user_prompt, bot_response in history:
            prompt += f"{user_prompt}\n"
            #print(prompt)
            prompt += f"{bot_response}\n"
            #print(prompt)
    prompt += f"<start_of_turn>user{message}<end_of_turn><start_of_turn>model"
    return prompt

def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem):
    #token max=8192
    hist_len=0
    client=clients[int(client_choice)-1]
    if not history:
        history = []
        hist_len=0
    if not memory:
        memory = []
        mem_len=0        
    if memory:
        for ea in memory[0-chat_mem:]:
            hist_len+=len(str(ea))
    in_len=len(system_prompt+prompt)+hist_len

    if (in_len+tokens) > 8000:
        history.append((prompt,"Wait, that's too many tokens, please reduce the 'Chat Memory' value, or reduce the 'Max new tokens' value"))
        yield history,memory
    else:
        generate_kwargs = dict(
            temperature=temp,
            max_new_tokens=tokens,
            top_p=top_p,
            repetition_penalty=rep_p,
            do_sample=True,
            seed=seed,
        )
        formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", memory[0-chat_mem:])
        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)],memory
        history.append((prompt,output))
        memory.append((prompt,output))
        yield history,memory
    if VERBOSE==True:
        print("\n######### HIST "+str(in_len))
        print("\n######### TOKENS "+str(tokens))        
        print("\n######### PROMPT "+str(len(formatted_prompt)))

def get_screenshot(chat: list,height=5000,width=600,chatblock=[],theme="light",wait=3000,header=True):
    print(chatblock)
    tog = 0
    if chatblock:
        tog = 3
    result = ss_client.predict(str(chat),height,width,chatblock,header,theme,wait,api_name="/run_script")
    out = f'https://omnibus-html-image-current-tab.hf.space/file={result[tog]}'
    print(out)
    return out

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:
    memory=gr.State()
    gr.HTML("""<center><h1 style='font-size:xx-large;'>Google Gemma Models</h1><br><h3>running on Huggingface Inference Client</h3><br><h7>EXPERIMENTAL""")
    chat_b = 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],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=1600,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)
                    chat_mem=gr.Number(label="Chat Memory", info="Number of previous chats to retain",value=4)
        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)

    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,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem],[chat_b,memory])
    go=btn.click(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem],[chat_b,memory])
    stop_btn.click(None,None,None,cancels=[go,im_go,chat_sub])
    clear_btn.click(clear_fn,None,[inp,sys_inp,chat_b,memory])
app.queue(default_concurrency_limit=10).launch()