Spaces:
Paused
Paused
File size: 9,675 Bytes
9e1636e bc08d80 9b436aa 9e1636e 982cf4e 9e1636e 982cf4e 9e1636e 8c9d1a2 09896d5 8c9d1a2 e9d94bd b54983b 5779868 5fb9f60 d8ca67b ee61882 3736c10 5fb9f60 3736c10 7ef6710 467715f 3242987 5fb9f60 7ef6710 f50ab84 6c702e3 9e1636e 982cf4e 72b2420 9e1636e 72b2420 9e1636e 43c5b06 9e1636e ed36b09 43c5b06 9e1636e 43c5b06 9e1636e 43c5b06 9e1636e 43c5b06 9e1636e 982cf4e d1cc2c0 982cf4e 06d3d78 982cf4e 9e1636e 982cf4e 1b39c53 982cf4e 9e1636e d6acc70 84120b5 a2623ed 84120b5 a2623ed 84120b5 f2b1918 84120b5 65b131b 84120b5 317583d d6acc70 0718219 317583d d6acc70 84120b5 9e1636e cf25f9b 9e1636e 84120b5 9e1636e d1cc2c0 9e1636e 84120b5 9e1636e 84120b5 9e1636e 84120b5 9e1636e f9856aa |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 |
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import gradio as gr
#from transformers import pipeline
import torch
from utils import *
from presets import *
from huggingface_hub import login
from transformers import LlamaForCausalLM, LlamaTokenizer
#antwort=""
######################################################################
#Modelle und Tokenizer
#Hugging Chat nutzen
# Create a chatbot connection
#chatbot = hugchat.ChatBot(cookie_path="cookies.json")
#Alternativ mit beliebigen Modellen:
base_model = "project-baize/baize-v2-7b" #load_8bit = False (in load_tokenizer_and_model)
#base_model = "MAGAer13/mPLUG-Owl" #load_8bit = False (in load_tokenizer_and_model)
#base_model = "alexkueck/li-tis-tuned-2" #load_8bit = False (in load_tokenizer_and_model)
#base_model = "TheBloke/airoboros-13B-HF" #load_8bit = False (in load_tokenizer_and_model)
#base_model = "EleutherAI/gpt-neo-1.3B" #load_8bit = False (in load_tokenizer_and_model)
#base_model = "TheBloke/airoboros-13B-HF" #load_8bit = True
#base_model = "TheBloke/vicuna-13B-1.1-HF" #load_8bit = ?
#following runs only on GPU upgrade
#base_model = "TheBloke/airoboros-65B-gpt4-1.3-GPTQ" #model_basename = "airoboros-65b-gpt4-1.3-GPTQ-4bit--1g.act.order"
#base_model = "lmsys/vicuna-13b-v1.3"
#base_model = "gpt2-xl" # options: ['gpt2', 'gpt2-medium', 'gpt2-large', 'gpt2-xl']
####################################
#Model und Tokenzier laden
tokenizer,model,device = load_tokenizer_and_model(base_model,False)
################################
#Alternativ: Model und Tokenizer für GPT2
#tokenizer,model,device = load_tokenizer_and_model_gpt2(base_model,False)
#Alternativ bloke gpt3 und4 - only with GPU upgarde
#tokenizer,model,device = load_tokenizer_and_model_bloke_gpt(base_model, "airoboros-65b-gpt4-1.3-GPTQ-4bit--1g.act.order")
#Alternativ Model und Tokenzier laden für Baize
#tokenizer,model,device = load_tokenizer_and_model_Baize(base_model,False)
########################################################################
#Chat KI nutzen, um Text zu generieren...
def predict(text,
chatbotGr,
history,
top_p,
temperature,
max_length_tokens,
max_context_length_tokens,):
if text=="":
yield chatbotGr,history,"Empty context."
return
try:
model
except:
yield [[text,"No Model Found"]],[],"No Model Found"
return
inputs = generate_prompt_with_history(text,history,tokenizer,max_length=max_context_length_tokens)
if inputs is None:
yield chatbotGr,history,"Input too long."
return
else:
prompt,inputs=inputs
begin_length = len(prompt)
input_ids = inputs["input_ids"][:,-max_context_length_tokens:].to(device)
torch.cuda.empty_cache()
#torch.no_grad() bedeutet, dass für die betreffenden tensoren keine Ableitungen berechnet werden bei der backpropagation
#hier soll das NN ja auch nicht geändert werden 8backprop ist nicht nötig), da es um interference-prompts geht!
with torch.no_grad():
#die vergangenen prompts werden alle als Tupel in history abgelegt sortiert nach 'Human' und 'AI'- dass sind daher auch die stop-words, die den jeweils nächsten Eintrag kennzeichnen
for x in greedy_search(input_ids,model,tokenizer,stop_words=["[|Human|]", "[|AI|]"],max_length=max_length_tokens,temperature=temperature,top_p=top_p):
if is_stop_word_or_prefix(x,["[|Human|]", "[|AI|]"]) is False:
if "[|Human|]" in x:
x = x[:x.index("[|Human|]")].strip()
if "[|AI|]" in x:
x = x[:x.index("[|AI|]")].strip()
x = x.strip()
a, b= [[y[0],convert_to_markdown(y[1])] for y in history]+[[text, convert_to_markdown(x)]],history + [[text,x]]
yield a, b, "Generating..."
if shared_state.interrupted:
shared_state.recover()
try:
yield a, b, "Stop: Success"
return
except:
pass
del input_ids
gc.collect()
torch.cuda.empty_cache()
try:
yield a,b,"Generate: Success"
except:
pass
def reset_chat():
#id_new = chatbot.new_conversation()
#chatbot.change_conversation(id_new)
reset_textbox()
##########################################################
#Übersetzungs Ki nutzen
def translate():
return "Kommt noch!"
#Programmcode KI
def coding():
return "Kommt noch!"
#######################################################################
#Darstellung mit Gradio
with open("custom.css", "r", encoding="utf-8") as f:
customCSS = f.read()
with gr.Blocks(theme=small_and_beautiful_theme) as demo:
history = gr.State([])
user_question = gr.State("")
gr.Markdown("KIs am LI - wähle aus, was du bzgl. KI-Bots ausprobieren möchtest!")
with gr.Tabs():
with gr.TabItem("LI-Chat"):
with gr.Row():
gr.HTML(title)
status_display = gr.Markdown("Erfolg", elem_id="status_display")
gr.Markdown(description_top)
with gr.Row(scale=1).style(equal_height=True):
with gr.Column(scale=5):
with gr.Row(scale=1):
chatbotGr = gr.Chatbot(elem_id="LI_chatbot").style(height="100%")
with gr.Row(scale=1):
with gr.Column(scale=12):
user_input = gr.Textbox(
show_label=False, placeholder="Gib deinen Text / Frage ein."
).style(container=False)
with gr.Column(min_width=100, scale=1):
submitBtn = gr.Button("Absenden")
with gr.Column(min_width=100, scale=1):
cancelBtn = gr.Button("Stoppen")
with gr.Row(scale=1):
emptyBtn = gr.Button(
"🧹 Neuer Chat",
)
with gr.Column():
with gr.Column(min_width=50, scale=1):
with gr.Tab(label="Nur zum Testen:"):
gr.Markdown("# Parameter")
top_p = gr.Slider(
minimum=-0,
maximum=1.0,
value=0.95,
step=0.05,
interactive=True,
label="Top-p",
)
temperature = gr.Slider(
minimum=0.1,
maximum=2.0,
value=1,
step=0.1,
interactive=True,
label="Temperature",
)
max_length_tokens = gr.Slider(
minimum=0,
maximum=512,
value=512,
step=8,
interactive=True,
label="Max Generation Tokens",
)
max_context_length_tokens = gr.Slider(
minimum=0,
maximum=4096,
value=2048,
step=128,
interactive=True,
label="Max History Tokens",
)
gr.Markdown(description)
with gr.TabItem("Übersetzungen"):
with gr.Row():
gr.Textbox(
show_label=False, placeholder="Ist noch in Arbeit..."
).style(container=False)
with gr.TabItem("Code-Generierungen"):
with gr.Row():
gr.Textbox(
show_label=False, placeholder="Ist noch in Arbeit..."
).style(container=False)
predict_args = dict(
fn=predict,
inputs=[
user_question,
chatbotGr,
history,
top_p,
temperature, #Variation der Antworten - stand. 1.0
max_length_tokens,
max_context_length_tokens,
],
outputs=[chatbotGr, history, status_display],
show_progress=True,
)
#neuer Chat
reset_args = dict(
#fn=reset_chat, inputs=[], outputs=[user_input, status_display]
fn=reset_textbox, inputs=[], outputs=[user_input, status_display]
)
# Chatbot
transfer_input_args = dict(
fn=transfer_input, inputs=[user_input], outputs=[user_question, user_input, submitBtn], show_progress=True
)
#Listener auf Start-Click auf Button oder Return
predict_event1 = user_input.submit(**transfer_input_args).then(**predict_args)
predict_event2 = submitBtn.click(**transfer_input_args).then(**predict_args)
#Listener, Wenn reset...
emptyBtn.click(
reset_state,
outputs=[chatbotGr, history, status_display],
show_progress=True,
)
emptyBtn.click(**reset_args)
demo.title = "LI Chat"
#demo.queue(concurrency_count=1).launch(share=True)
demo.queue(concurrency_count=1).launch(debug=True)
|