wissemkarous commited on
Commit
8616e4c
1 Parent(s): dc60236
Files changed (1) hide show
  1. app.py +160 -0
app.py ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ from gradio_client import Client
3
+ from huggingface_hub import InferenceClient
4
+ import random
5
+ ss_client = Client("https://omnibus-html-image-current-tab.hf.space/")
6
+
7
+ models=[
8
+ "google/gemma-7b",
9
+ "google/gemma-7b-it",
10
+ "google/gemma-2b",
11
+ "google/gemma-2b-it"
12
+ ]
13
+ clients=[
14
+ InferenceClient(models[0]),
15
+ InferenceClient(models[1]),
16
+ InferenceClient(models[2]),
17
+ InferenceClient(models[3]),
18
+ ]
19
+
20
+ VERBOSE=False
21
+
22
+ def load_models(inp):
23
+ if VERBOSE==True:
24
+ print(type(inp))
25
+ print(inp)
26
+ print(models[inp])
27
+ #client_z.clear()
28
+ #client_z.append(InferenceClient(models[inp]))
29
+ return gr.update(label=models[inp])
30
+
31
+ def format_prompt(message, history, cust_p):
32
+ prompt = ""
33
+ if history:
34
+ for user_prompt, bot_response in history:
35
+ prompt += f"<start_of_turn>user{user_prompt}<end_of_turn>"
36
+ prompt += f"<start_of_turn>model{bot_response}<end_of_turn>"
37
+ if VERBOSE==True:
38
+ print(prompt)
39
+ #prompt += f"<start_of_turn>user\n{message}<end_of_turn>\n<start_of_turn>model\n"
40
+ prompt+=cust_p.replace("USER_INPUT",message)
41
+ return prompt
42
+
43
+ def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem,cust_p):
44
+ #token max=8192
45
+ print(client_choice)
46
+ hist_len=0
47
+ client=clients[int(client_choice)-1]
48
+ if not history:
49
+ history = []
50
+ hist_len=0
51
+ if not memory:
52
+ memory = []
53
+ mem_len=0
54
+ if memory:
55
+ for ea in memory[0-chat_mem:]:
56
+ hist_len+=len(str(ea))
57
+ in_len=len(system_prompt+prompt)+hist_len
58
+
59
+ if (in_len+tokens) > 8000:
60
+ history.append((prompt,"Wait, that's too many tokens, please reduce the 'Chat Memory' value, or reduce the 'Max new tokens' value"))
61
+ yield history,memory
62
+ else:
63
+ generate_kwargs = dict(
64
+ temperature=temp,
65
+ max_new_tokens=tokens,
66
+ top_p=top_p,
67
+ repetition_penalty=rep_p,
68
+ do_sample=True,
69
+ seed=seed,
70
+ )
71
+ if system_prompt:
72
+ formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", memory[0-chat_mem:],cust_p)
73
+ else:
74
+ formatted_prompt = format_prompt(prompt, memory[0-chat_mem:],cust_p)
75
+ stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
76
+ output = ""
77
+ for response in stream:
78
+ output += response.token.text
79
+ yield [(prompt,output)],memory
80
+ history.append((prompt,output))
81
+ memory.append((prompt,output))
82
+ yield history,memory
83
+
84
+ if VERBOSE==True:
85
+ print("\n######### HIST "+str(in_len))
86
+ print("\n######### TOKENS "+str(tokens))
87
+
88
+ def get_screenshot(chat: list,height=5000,width=600,chatblock=[],theme="light",wait=3000,header=True):
89
+ print(chatblock)
90
+ tog = 0
91
+ if chatblock:
92
+ tog = 3
93
+ result = ss_client.predict(str(chat),height,width,chatblock,header,theme,wait,api_name="/run_script")
94
+ out = f'https://omnibus-html-image-current-tab.hf.space/file={result[tog]}'
95
+ print(out)
96
+ return out
97
+
98
+ def clear_fn():
99
+ return None,None,None,None
100
+ rand_val=random.randint(1,1111111111111111)
101
+
102
+ def check_rand(inp,val):
103
+ if inp==True:
104
+ return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1,1111111111111111))
105
+ else:
106
+ return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val))
107
+
108
+ with gr.Blocks() as app:
109
+ memory=gr.State()
110
+ gr.HTML("""<center><h1 style='font-size:xx-large;'>Google Gemma Models</h1><br><h3>running on Huggingface Inference Client</h3><br><h7>EXPERIMENTAL""")
111
+ chat_b = gr.Chatbot(height=500)
112
+ with gr.Group():
113
+ with gr.Row():
114
+ with gr.Column(scale=3):
115
+ inp = gr.Textbox(label="Prompt")
116
+ sys_inp = gr.Textbox(label="System Prompt (optional)")
117
+ with gr.Accordion("Prompt Format",open=False):
118
+ custom_prompt=gr.Textbox(label="Modify Prompt Format", info="For testing purposes. 'USER_INPUT' is where 'SYSTEM_PROMPT, PROMPT' will be placed", lines=3,value="<start_of_turn>userUSER_INPUT<end_of_turn><start_of_turn>model")
119
+ with gr.Row():
120
+ with gr.Column(scale=2):
121
+ btn = gr.Button("Chat")
122
+ with gr.Column(scale=1):
123
+ with gr.Group():
124
+ stop_btn=gr.Button("Stop")
125
+ clear_btn=gr.Button("Clear")
126
+ client_choice=gr.Dropdown(label="Models",type='index',choices=[c for c in models],value=models[0],interactive=True)
127
+ with gr.Column(scale=1):
128
+ with gr.Group():
129
+ rand = gr.Checkbox(label="Random Seed", value=True)
130
+ seed=gr.Slider(label="Seed", minimum=1, maximum=1111111111111111,step=1, value=rand_val)
131
+ 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")
132
+ temp=gr.Slider(label="Temperature",step=0.01, minimum=0.01, maximum=1.0, value=0.49)
133
+ top_p=gr.Slider(label="Top-P",step=0.01, minimum=0.01, maximum=1.0, value=0.49)
134
+ rep_p=gr.Slider(label="Repetition Penalty",step=0.01, minimum=0.1, maximum=2.0, value=0.99)
135
+ chat_mem=gr.Number(label="Chat Memory", info="Number of previous chats to retain",value=4)
136
+ with gr.Accordion(label="Screenshot",open=False):
137
+ with gr.Row():
138
+ with gr.Column(scale=3):
139
+ im_btn=gr.Button("Screenshot")
140
+ img=gr.Image(type='filepath')
141
+ with gr.Column(scale=1):
142
+ with gr.Row():
143
+ im_height=gr.Number(label="Height",value=5000)
144
+ im_width=gr.Number(label="Width",value=500)
145
+ wait_time=gr.Number(label="Wait Time",value=3000)
146
+ theme=gr.Radio(label="Theme", choices=["light","dark"],value="light")
147
+ chatblock=gr.Dropdown(label="Chatblocks",info="Choose specific blocks of chat",choices=[c for c in range(1,40)],multiselect=True)
148
+
149
+
150
+ client_choice.change(load_models,client_choice,[chat_b])
151
+ app.load(load_models,client_choice,[chat_b])
152
+
153
+ im_go=im_btn.click(get_screenshot,[chat_b,im_height,im_width,chatblock,theme,wait_time],img)
154
+
155
+ 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,custom_prompt],[chat_b,memory])
156
+ 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,custom_prompt],[chat_b,memory])
157
+
158
+ stop_btn.click(None,None,None,cancels=[go,im_go,chat_sub])
159
+ clear_btn.click(clear_fn,None,[inp,sys_inp,chat_b,memory])
160
+ app.queue(default_concurrency_limit=10).launch()