Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
+
import torch
|
4 |
+
import spaces
|
5 |
+
|
6 |
+
# Load the model and tokenizer from Hugging Face
|
7 |
+
model_path = "Ozaii/W.AI-13B-Chat" # Replace with your username and repository name
|
8 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
9 |
+
model = AutoModelForCausalLM.from_pretrained(model_path)
|
10 |
+
|
11 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
12 |
+
model.to(device)
|
13 |
+
model.eval()
|
14 |
+
|
15 |
+
@spaces.GPU
|
16 |
+
def generate_response(user_input, chat_history):
|
17 |
+
max_context_length = 750
|
18 |
+
max_response_length = 150
|
19 |
+
|
20 |
+
prompt = ""
|
21 |
+
for message in chat_history:
|
22 |
+
if message[0] is not None:
|
23 |
+
prompt += f"User: {message[0]}\n"
|
24 |
+
if message[1] is not None:
|
25 |
+
prompt += f"Assistant: {message[1]}\n"
|
26 |
+
prompt += f"***Use natural conversation style of assistant, you can use +18 swear words and use irregular language similar to assistant.***User: {user_input}\nAssistant:"
|
27 |
+
|
28 |
+
prompt_tokens = tokenizer.encode(prompt, add_special_tokens=False)
|
29 |
+
if len(prompt_tokens) > max_context_length:
|
30 |
+
prompt_tokens = prompt_tokens[-max_context_length:]
|
31 |
+
prompt = tokenizer.decode(prompt_tokens, clean_up_tokenization_spaces=True)
|
32 |
+
|
33 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(device)
|
34 |
+
with torch.no_grad():
|
35 |
+
outputs = model.generate(
|
36 |
+
inputs.input_ids,
|
37 |
+
max_length=len(inputs.input_ids[0]) + max_response_length, # Limit the maximum length for context and response
|
38 |
+
min_length = 45,
|
39 |
+
temperature=0.7, # Slightly higher temperature for more diverse responses
|
40 |
+
top_k=30,
|
41 |
+
top_p=0.9, # Allow a bit more randomness
|
42 |
+
repetition_penalty=1.1, # Mild repetition penalty
|
43 |
+
no_repeat_ngram_size=3, # Ensure no repeated phrases
|
44 |
+
eos_token_id=tokenizer.eos_token_id,
|
45 |
+
pad_token_id=tokenizer.eos_token_id
|
46 |
+
)
|
47 |
+
|
48 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
49 |
+
assistant_response = response.split("Assistant:")[-1].strip()
|
50 |
+
assistant_response = assistant_response.split('\n')[0].strip()
|
51 |
+
chat_history.append((user_input, assistant_response))
|
52 |
+
return chat_history, chat_history
|
53 |
+
|
54 |
+
def restart_chat():
|
55 |
+
return [], []
|
56 |
+
|
57 |
+
with gr.Blocks() as chat_interface:
|
58 |
+
gr.Markdown("<h1><center>W.AI Chat Nikker xD</center></h1>")
|
59 |
+
chat_history = gr.State([])
|
60 |
+
with gr.Column():
|
61 |
+
chatbox = gr.Chatbot()
|
62 |
+
with gr.Row():
|
63 |
+
user_input = gr.Textbox(show_label=False, placeholder="Summon Wali Here...")
|
64 |
+
submit_button = gr.Button("Send")
|
65 |
+
restart_button = gr.Button("Restart")
|
66 |
+
|
67 |
+
submit_button.click(
|
68 |
+
generate_response,
|
69 |
+
inputs=[user_input, chat_history],
|
70 |
+
outputs=[chatbox, chat_history]
|
71 |
+
)
|
72 |
+
|
73 |
+
restart_button.click(
|
74 |
+
restart_chat,
|
75 |
+
inputs=[],
|
76 |
+
outputs=[chatbox, chat_history]
|
77 |
+
)
|
78 |
+
|
79 |
+
chat_interface.launch()
|