Spaces:
Sleeping
Sleeping
Create Test.py
Browse files
Test.py
ADDED
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
import spaces
|
5 |
+
|
6 |
+
model_name = "Sakalti/SakalFusion-7B-Alpha"
|
7 |
+
|
8 |
+
model = AutoModelForCausalLM.from_pretrained(
|
9 |
+
model_name,
|
10 |
+
torch_dtype=torch.bfloat16,
|
11 |
+
device_map="auto"
|
12 |
+
)
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
14 |
+
|
15 |
+
@spaces.gpu(duration=100)
|
16 |
+
def generate(prompt, history, top_p, top_k, max_new_tokens, repetition_penalty, temperature):
|
17 |
+
messages = [
|
18 |
+
{"role": "system", "content": "γγͺγγ―γγ¬γ³γγͺγΌγͺγγ£γγγγγγ§γγ"},
|
19 |
+
{"role": "user", "content": prompt}
|
20 |
+
]
|
21 |
+
text = tokenizer.apply_chat_template(
|
22 |
+
messages,
|
23 |
+
tokenize=False,
|
24 |
+
add_generation_prompt=True
|
25 |
+
)
|
26 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
27 |
+
|
28 |
+
generated_ids = model.generate(
|
29 |
+
**model_inputs,
|
30 |
+
max_new_tokens=max_new_tokens,
|
31 |
+
top_p=top_p,
|
32 |
+
top_k=top_k,
|
33 |
+
repetition_penalty=repetition_penalty,
|
34 |
+
temperature=temperature
|
35 |
+
)
|
36 |
+
generated_ids = [
|
37 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
38 |
+
]
|
39 |
+
|
40 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
41 |
+
return response, history + [[prompt, response]]
|
42 |
+
|
43 |
+
with gr.Blocks() as demo:
|
44 |
+
chatbot = gr.Chatbot()
|
45 |
+
msg = gr.Textbox()
|
46 |
+
clear = gr.Button("Clear")
|
47 |
+
|
48 |
+
with gr.Row():
|
49 |
+
top_p = gr.Slider(0.0, 1.0, value=0.9, label="Top P")
|
50 |
+
top_k = gr.Slider(0, 100, value=50, label="Top K")
|
51 |
+
max_new_tokens = gr.Slider(1, 2048, value=864, label="Max New Tokens")
|
52 |
+
repetition_penalty = gr.Slider(1.0, 2.0, value=1.2, label="Repetition Penalty")
|
53 |
+
temperature = gr.Slider(0.1, 1.0, value=0.7, label="Temperature")
|
54 |
+
|
55 |
+
def respond(message, chat_history, top_p, top_k, max_new_tokens, repetition_penalty, temperature):
|
56 |
+
bot_message, chat_history = generate(message, chat_history, top_p, top_k, max_new_tokens, repetition_penalty, temperature)
|
57 |
+
return "", chat_history, chat_history
|
58 |
+
|
59 |
+
msg.submit(respond, [msg, chatbot, top_p, top_k, max_new_tokens, repetition_penalty, temperature], [msg, chatbot, chatbot])
|
60 |
+
clear.click(lambda: ([], []), None, [chatbot, msg])
|
61 |
+
|
62 |
+
demo.launch(share=True)
|