update file and model for only 1bit
Browse files
app.py
CHANGED
|
@@ -1,3 +1,296 @@
|
|
| 1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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 |
+
|
| 9 |
+
"1bitLLM/bitnet_b1_58-3B",
|
| 10 |
+
"1bitLLM/bitnet_b1_58-large",
|
| 11 |
+
"1bitLLM/bitnet_b1_58-xl",
|
| 12 |
+
]
|
| 13 |
+
client_z=[]
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
def load_models(inp,new_models):
|
| 17 |
+
if not new_models:
|
| 18 |
+
new_models=models
|
| 19 |
+
out_box=[gr.Chatbot(),gr.Chatbot(),gr.Chatbot(),gr.Chatbot()]
|
| 20 |
+
print(type(inp))
|
| 21 |
+
print(inp)
|
| 22 |
+
#print(new_models[inp[0]])
|
| 23 |
+
client_z.clear()
|
| 24 |
+
for z,ea in enumerate(inp):
|
| 25 |
+
client_z.append(InferenceClient(new_models[inp[z]]))
|
| 26 |
+
out_box[z]=(gr.update(label=new_models[inp[z]]))
|
| 27 |
+
return out_box[0],out_box[1],out_box[2],out_box[3]
|
| 28 |
+
|
| 29 |
+
def format_prompt_default(message, history):
|
| 30 |
+
prompt = ""
|
| 31 |
+
if history:
|
| 32 |
+
#<start_of_turn>userHow does the brain work?<end_of_turn><start_of_turn>model
|
| 33 |
+
for user_prompt, bot_response in history:
|
| 34 |
+
prompt += f"{user_prompt}\n"
|
| 35 |
+
print(prompt)
|
| 36 |
+
prompt += f"{bot_response}\n"
|
| 37 |
+
print(prompt)
|
| 38 |
+
prompt += f"{message}\n"
|
| 39 |
+
return prompt
|
| 40 |
+
|
| 41 |
+
def format_prompt_gemma(message, history):
|
| 42 |
+
prompt = ""
|
| 43 |
+
if history:
|
| 44 |
+
#<start_of_turn>userHow does the brain work?<end_of_turn><start_of_turn>model
|
| 45 |
+
for user_prompt, bot_response in history:
|
| 46 |
+
prompt += f"{user_prompt}\n"
|
| 47 |
+
print(prompt)
|
| 48 |
+
prompt += f"{bot_response}\n"
|
| 49 |
+
print(prompt)
|
| 50 |
+
prompt += f"<start_of_turn>user{message}<end_of_turn><start_of_turn>model"
|
| 51 |
+
return prompt
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
def format_prompt_mixtral(message, history):
|
| 55 |
+
prompt = "<s>"
|
| 56 |
+
if history:
|
| 57 |
+
for user_prompt, bot_response in history:
|
| 58 |
+
prompt += f"[INST] {user_prompt} [/INST]"
|
| 59 |
+
prompt += f" {bot_response}</s> "
|
| 60 |
+
prompt += f"[INST] {message} [/INST]"
|
| 61 |
+
return prompt
|
| 62 |
+
|
| 63 |
+
def format_prompt_choose(message, history, model_name, new_models=None):
|
| 64 |
+
if not new_models:
|
| 65 |
+
new_models=models
|
| 66 |
+
if "gemma" in new_models[model_name].lower() and "it" in new_models[model_name].lower():
|
| 67 |
+
return format_prompt_gemma(message,history)
|
| 68 |
+
if "mixtral" in new_models[model_name].lower():
|
| 69 |
+
return format_prompt_mixtral(message,history)
|
| 70 |
+
else:
|
| 71 |
+
return format_prompt_default(message,history)
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
mega_hist=[[],[],[],[]]
|
| 76 |
+
def chat_inf_tree(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p,hid_val):
|
| 77 |
+
if len(client_choice)>=hid_val:
|
| 78 |
+
client=client_z[int(hid_val)-1]
|
| 79 |
+
if history:
|
| 80 |
+
mega_hist[hid_val-1]=history
|
| 81 |
+
#history = []
|
| 82 |
+
hist_len=0
|
| 83 |
+
generate_kwargs = dict(
|
| 84 |
+
temperature=temp,
|
| 85 |
+
max_new_tokens=tokens,
|
| 86 |
+
top_p=top_p,
|
| 87 |
+
repetition_penalty=rep_p,
|
| 88 |
+
do_sample=True,
|
| 89 |
+
seed=seed,
|
| 90 |
+
)
|
| 91 |
+
#formatted_prompt=prompt
|
| 92 |
+
formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", mega_hist[hid_val-1])
|
| 93 |
+
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
| 94 |
+
output = ""
|
| 95 |
+
for response in stream:
|
| 96 |
+
output += response.token.text
|
| 97 |
+
yield [(prompt,output)]
|
| 98 |
+
mega_hist[hid_val-1].append((prompt,output))
|
| 99 |
+
yield mega_hist[hid_val-1]
|
| 100 |
+
else:
|
| 101 |
+
yield None
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
|
| 105 |
+
|
| 106 |
+
def chat_inf_a(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p,hid_val):
|
| 107 |
+
if len(client_choice)>=hid_val:
|
| 108 |
+
if system_prompt:
|
| 109 |
+
system_prompt=f'{system_prompt}, '
|
| 110 |
+
client1=client_z[int(hid_val)-1]
|
| 111 |
+
if not history:
|
| 112 |
+
history = []
|
| 113 |
+
hist_len=0
|
| 114 |
+
generate_kwargs = dict(
|
| 115 |
+
temperature=temp,
|
| 116 |
+
max_new_tokens=tokens,
|
| 117 |
+
top_p=top_p,
|
| 118 |
+
repetition_penalty=rep_p,
|
| 119 |
+
do_sample=True,
|
| 120 |
+
seed=seed,
|
| 121 |
+
)
|
| 122 |
+
#formatted_prompt=prompt
|
| 123 |
+
formatted_prompt = format_prompt_choose(f"{system_prompt}{prompt}", history, client_choice[0])
|
| 124 |
+
stream1 = client1.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
| 125 |
+
output = ""
|
| 126 |
+
for response in stream1:
|
| 127 |
+
output += response.token.text
|
| 128 |
+
yield [(prompt,output)]
|
| 129 |
+
history.append((prompt,output))
|
| 130 |
+
yield history
|
| 131 |
+
else:
|
| 132 |
+
yield None
|
| 133 |
+
|
| 134 |
+
|
| 135 |
+
def chat_inf_b(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p,hid_val):
|
| 136 |
+
if len(client_choice)>=hid_val:
|
| 137 |
+
if system_prompt:
|
| 138 |
+
system_prompt=f'{system_prompt}, '
|
| 139 |
+
client2=client_z[int(hid_val)-1]
|
| 140 |
+
if not history:
|
| 141 |
+
history = []
|
| 142 |
+
hist_len=0
|
| 143 |
+
generate_kwargs = dict(
|
| 144 |
+
temperature=temp,
|
| 145 |
+
max_new_tokens=tokens,
|
| 146 |
+
top_p=top_p,
|
| 147 |
+
repetition_penalty=rep_p,
|
| 148 |
+
do_sample=True,
|
| 149 |
+
seed=seed,
|
| 150 |
+
)
|
| 151 |
+
#formatted_prompt=prompt
|
| 152 |
+
formatted_prompt = format_prompt_choose(f"{system_prompt}{prompt}", history, client_choice[1])
|
| 153 |
+
stream2 = client2.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
| 154 |
+
output = ""
|
| 155 |
+
for response in stream2:
|
| 156 |
+
output += response.token.text
|
| 157 |
+
yield [(prompt,output)]
|
| 158 |
+
history.append((prompt,output))
|
| 159 |
+
yield history
|
| 160 |
+
else:
|
| 161 |
+
yield None
|
| 162 |
+
|
| 163 |
+
def chat_inf_c(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p,hid_val):
|
| 164 |
+
if len(client_choice)>=hid_val:
|
| 165 |
+
if system_prompt:
|
| 166 |
+
system_prompt=f'{system_prompt}, '
|
| 167 |
+
client3=client_z[int(hid_val)-1]
|
| 168 |
+
if not history:
|
| 169 |
+
history = []
|
| 170 |
+
hist_len=0
|
| 171 |
+
generate_kwargs = dict(
|
| 172 |
+
temperature=temp,
|
| 173 |
+
max_new_tokens=tokens,
|
| 174 |
+
top_p=top_p,
|
| 175 |
+
repetition_penalty=rep_p,
|
| 176 |
+
do_sample=True,
|
| 177 |
+
seed=seed,
|
| 178 |
+
)
|
| 179 |
+
#formatted_prompt=prompt
|
| 180 |
+
formatted_prompt = format_prompt_choose(f"{system_prompt}{prompt}", history, client_choice[2])
|
| 181 |
+
stream3 = client3.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
| 182 |
+
output = ""
|
| 183 |
+
for response in stream3:
|
| 184 |
+
output += response.token.text
|
| 185 |
+
yield [(prompt,output)]
|
| 186 |
+
history.append((prompt,output))
|
| 187 |
+
yield history
|
| 188 |
+
else:
|
| 189 |
+
yield None
|
| 190 |
+
|
| 191 |
+
def chat_inf_d(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p,hid_val):
|
| 192 |
+
if len(client_choice)>=hid_val:
|
| 193 |
+
if system_prompt:
|
| 194 |
+
system_prompt=f'{system_prompt}, '
|
| 195 |
+
client4=client_z[int(hid_val)-1]
|
| 196 |
+
if not history:
|
| 197 |
+
history = []
|
| 198 |
+
hist_len=0
|
| 199 |
+
generate_kwargs = dict(
|
| 200 |
+
temperature=temp,
|
| 201 |
+
max_new_tokens=tokens,
|
| 202 |
+
top_p=top_p,
|
| 203 |
+
repetition_penalty=rep_p,
|
| 204 |
+
do_sample=True,
|
| 205 |
+
seed=seed,
|
| 206 |
+
)
|
| 207 |
+
#formatted_prompt=prompt
|
| 208 |
+
formatted_prompt = format_prompt_choose(f"{system_prompt}{prompt}", history, client_choice[3])
|
| 209 |
+
stream4 = client4.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
| 210 |
+
output = ""
|
| 211 |
+
for response in stream4:
|
| 212 |
+
output += response.token.text
|
| 213 |
+
yield [(prompt,output)]
|
| 214 |
+
history.append((prompt,output))
|
| 215 |
+
yield history
|
| 216 |
+
else:
|
| 217 |
+
yield None
|
| 218 |
+
def add_new_model(inp, cur):
|
| 219 |
+
cur.append(inp)
|
| 220 |
+
return cur,gr.update(choices=[z for z in cur])
|
| 221 |
+
def load_new(models=models):
|
| 222 |
+
return models
|
| 223 |
+
|
| 224 |
+
def clear_fn():
|
| 225 |
+
return None,None,None,None,None,None
|
| 226 |
+
rand_val=random.randint(1,1111111111111111)
|
| 227 |
+
def check_rand(inp,val):
|
| 228 |
+
if inp==True:
|
| 229 |
+
return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1,1111111111111111))
|
| 230 |
+
else:
|
| 231 |
+
return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val))
|
| 232 |
+
|
| 233 |
+
with gr.Blocks() as app:
|
| 234 |
+
new_models=gr.State([])
|
| 235 |
+
gr.HTML("""<center><h1 style='font-size:xx-large;'>Chatbot Model Compare</h1><br><h3>running on Huggingface Inference Client</h3><br><h7>EXPERIMENTAL""")
|
| 236 |
+
with gr.Row():
|
| 237 |
+
chat_a = gr.Chatbot(height=500)
|
| 238 |
+
chat_b = gr.Chatbot(height=500)
|
| 239 |
+
with gr.Row():
|
| 240 |
+
chat_c = gr.Chatbot(height=500)
|
| 241 |
+
chat_d = gr.Chatbot(height=500)
|
| 242 |
+
with gr.Group():
|
| 243 |
+
with gr.Row():
|
| 244 |
+
with gr.Column(scale=3):
|
| 245 |
+
inp = gr.Textbox(label="Prompt")
|
| 246 |
+
sys_inp = gr.Textbox(label="System Prompt (optional)")
|
| 247 |
+
with gr.Row():
|
| 248 |
+
with gr.Column(scale=2):
|
| 249 |
+
btn = gr.Button("Chat")
|
| 250 |
+
with gr.Column(scale=1):
|
| 251 |
+
with gr.Group():
|
| 252 |
+
stop_btn=gr.Button("Stop")
|
| 253 |
+
clear_btn=gr.Button("Clear")
|
| 254 |
+
client_choice=gr.Dropdown(label="Models",type='index',choices=[c for c in models],max_choices=4,multiselect=True,interactive=True)
|
| 255 |
+
add_model=gr.Textbox(label="New Model")
|
| 256 |
+
add_btn=gr.Button("Add Model")
|
| 257 |
+
with gr.Column(scale=1):
|
| 258 |
+
with gr.Group():
|
| 259 |
+
rand = gr.Checkbox(label="Random Seed", value=True)
|
| 260 |
+
seed=gr.Slider(label="Seed", minimum=1, maximum=1111111111111111,step=1, value=rand_val)
|
| 261 |
+
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")
|
| 262 |
+
temp=gr.Slider(label="Temperature",step=0.01, minimum=0.01, maximum=1.0, value=0.9)
|
| 263 |
+
top_p=gr.Slider(label="Top-P",step=0.01, minimum=0.01, maximum=1.0, value=0.9)
|
| 264 |
+
rep_p=gr.Slider(label="Repetition Penalty",step=0.1, minimum=0.1, maximum=2.0, value=1.0)
|
| 265 |
+
with gr.Accordion(label="Screenshot",open=False):
|
| 266 |
+
with gr.Row():
|
| 267 |
+
with gr.Column(scale=3):
|
| 268 |
+
im_btn=gr.Button("Screenshot")
|
| 269 |
+
img=gr.Image(type='filepath')
|
| 270 |
+
with gr.Column(scale=1):
|
| 271 |
+
with gr.Row():
|
| 272 |
+
im_height=gr.Number(label="Height",value=5000)
|
| 273 |
+
im_width=gr.Number(label="Width",value=500)
|
| 274 |
+
wait_time=gr.Number(label="Wait Time",value=3000)
|
| 275 |
+
theme=gr.Radio(label="Theme", choices=["light","dark"],value="light")
|
| 276 |
+
chatblock=gr.Dropdown(label="Chatblocks",info="Choose specific blocks of chat",choices=[c for c in range(1,40)],multiselect=True)
|
| 277 |
+
hid1=gr.Number(value=1,visible=False)
|
| 278 |
+
hid2=gr.Number(value=2,visible=False)
|
| 279 |
+
hid3=gr.Number(value=3,visible=False)
|
| 280 |
+
hid4=gr.Number(value=4,visible=False)
|
| 281 |
+
|
| 282 |
+
app.load(load_new,None,new_models)
|
| 283 |
+
add_btn.click(add_new_model,[add_model,new_models],[new_models,client_choice])
|
| 284 |
+
client_choice.change(load_models,[client_choice,new_models],[chat_a,chat_b,chat_c,chat_d])
|
| 285 |
+
|
| 286 |
+
#im_go=im_btn.click(get_screenshot,[chat_b,im_height,im_width,chatblock,theme,wait_time],img)
|
| 287 |
+
#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)
|
| 288 |
+
|
| 289 |
+
go1=btn.click(check_rand,[rand,seed],seed).then(chat_inf_a,[sys_inp,inp,chat_b,client_choice,seed,temp,tokens,top_p,rep_p,hid1],chat_a)
|
| 290 |
+
go2=btn.click(check_rand,[rand,seed],seed).then(chat_inf_b,[sys_inp,inp,chat_b,client_choice,seed,temp,tokens,top_p,rep_p,hid2],chat_b)
|
| 291 |
+
go3=btn.click(check_rand,[rand,seed],seed).then(chat_inf_c,[sys_inp,inp,chat_b,client_choice,seed,temp,tokens,top_p,rep_p,hid3],chat_c)
|
| 292 |
+
go4=btn.click(check_rand,[rand,seed],seed).then(chat_inf_d,[sys_inp,inp,chat_b,client_choice,seed,temp,tokens,top_p,rep_p,hid4],chat_d)
|
| 293 |
+
|
| 294 |
+
stop_btn.click(None,None,None,cancels=[go1,go2,go3,go4])
|
| 295 |
+
clear_btn.click(clear_fn,None,[inp,sys_inp,chat_a,chat_b,chat_c,chat_d])
|
| 296 |
+
app.queue(default_concurrency_limit=10).launch()
|