Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,117 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import (
|
4 |
+
AutoModelForCausalLM,
|
5 |
+
AutoTokenizer,
|
6 |
+
TextIteratorStreamer,
|
7 |
+
)
|
8 |
+
import os
|
9 |
+
from threading import Thread
|
10 |
+
import spaces
|
11 |
+
import time
|
12 |
+
import subprocess
|
13 |
+
|
14 |
+
PLACEHOLDER = """
|
15 |
+
<div style="padding: 40px; text-align: center; display: flex; flex-direction: column; align-items: center;">
|
16 |
+
<img src="https://i.imgur.com/dgSNbTl.jpg" style="width: 90%; max-width: 650px; height: auto; opacity: 0.8; ">
|
17 |
+
<h1 style="font-size: 28px; margin-top: 20px; margin-bottom: 2px; opacity: 0.55;">mii-llm / Maestrale</h1>
|
18 |
+
<p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">Chiedi pure...</p>
|
19 |
+
</div>
|
20 |
+
"""
|
21 |
+
|
22 |
+
DESCRIPTION = """<div>
|
23 |
+
<p>🇮🇹 Italian LLM <a href="https://huggingface.co/mii-llm/maestrale-chat-v0.3-beta"><b>Maestrale Chat v0.3 beta</b></a>. Maestrale is a powerful language model for Italian, trained by mii-llm, based on Mistral 7B.</p>
|
24 |
+
<p>🔎 For more details about the Maestrale model and how to use it with <code>transformers</code>, visit the <a href="https://huggingface.co/mii-llm/maestrale-chat-v0.3-beta">model card</a>.</p>
|
25 |
+
</div>"""
|
26 |
+
|
27 |
+
tokenizer = AutoTokenizer.from_pretrained("mii-llm/maestrale-chat-v0.4-alpha")
|
28 |
+
model = AutoModelForCausalLM.from_pretrained("mii-llm/maestrale-chat-v0.4-alpha", device_map="auto")
|
29 |
+
|
30 |
+
terminators = [
|
31 |
+
tokenizer.eos_token_id,
|
32 |
+
tokenizer.convert_tokens_to_ids("<|im_end|>")
|
33 |
+
]
|
34 |
+
|
35 |
+
if torch.cuda.is_available():
|
36 |
+
device = torch.device("cuda")
|
37 |
+
print(f"Using GPU: {torch.cuda.get_device_name(device)}")
|
38 |
+
else:
|
39 |
+
device = torch.device("cpu")
|
40 |
+
print("Using CPU")
|
41 |
+
|
42 |
+
|
43 |
+
model = model.to(device)
|
44 |
+
|
45 |
+
|
46 |
+
@spaces.GPU(duration=60)
|
47 |
+
def chat(message, history, system, temperature, do_sample, max_tokens):
|
48 |
+
chat = [{"role": "system", "content": system}] if system else []
|
49 |
+
chat.extend(
|
50 |
+
{"role": role, "content": content}
|
51 |
+
for user, assistant in history
|
52 |
+
for role, content in [("user", user), ("assistant", assistant)]
|
53 |
+
)
|
54 |
+
chat.append({"role": "user", "content": message})
|
55 |
+
|
56 |
+
messages = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
|
57 |
+
model_inputs = tokenizer([messages], return_tensors="pt").to(device)
|
58 |
+
|
59 |
+
streamer = TextIteratorStreamer(
|
60 |
+
tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True
|
61 |
+
)
|
62 |
+
|
63 |
+
generate_kwargs = {
|
64 |
+
**model_inputs,
|
65 |
+
"streamer": streamer,
|
66 |
+
"max_new_tokens": max_tokens,
|
67 |
+
"do_sample": do_sample,
|
68 |
+
"temperature": temperature,
|
69 |
+
"eos_token_id": terminators,
|
70 |
+
}
|
71 |
+
|
72 |
+
thread = Thread(target=model.generate, kwargs=generate_kwargs)
|
73 |
+
thread.start()
|
74 |
+
|
75 |
+
partial_text = ""
|
76 |
+
for new_text in streamer:
|
77 |
+
partial_text += new_text
|
78 |
+
yield partial_text
|
79 |
+
|
80 |
+
yield partial_text
|
81 |
+
|
82 |
+
|
83 |
+
chatbot = gr.Chatbot(height=550, placeholder=PLACEHOLDER, label='Conversazione')
|
84 |
+
|
85 |
+
demo = gr.ChatInterface(
|
86 |
+
fn=chat,
|
87 |
+
chatbot=chatbot,
|
88 |
+
fill_height=True,
|
89 |
+
theme=gr.themes.Soft(),
|
90 |
+
additional_inputs_accordion=gr.Accordion(
|
91 |
+
label="⚙️ Parametri", open=False, render=False
|
92 |
+
),
|
93 |
+
additional_inputs=[
|
94 |
+
gr.Textbox(
|
95 |
+
label="System",
|
96 |
+
value="sei un assistente utile.",
|
97 |
+
),
|
98 |
+
gr.Slider(
|
99 |
+
minimum=0, maximum=1, step=0.1, value=0.7, label="Temperature", render=False
|
100 |
+
),
|
101 |
+
gr.Checkbox(label="Sampling", value=True),
|
102 |
+
gr.Slider(
|
103 |
+
minimum=128,
|
104 |
+
maximum=4096,
|
105 |
+
step=1,
|
106 |
+
value=512,
|
107 |
+
label="Max new tokens",
|
108 |
+
render=False,
|
109 |
+
),
|
110 |
+
],
|
111 |
+
stop_btn="Stop Generation",
|
112 |
+
cache_examples=False,
|
113 |
+
title="Maestrale Chat v0.3 beta",
|
114 |
+
description=DESCRIPTION
|
115 |
+
)
|
116 |
+
|
117 |
+
demo.launch()
|