Spaces:
Sleeping
Sleeping
yvoievid
commited on
Commit
•
f5242b6
1
Parent(s):
2688f05
Add simple prompting UI with pretrained LLama 2
Browse files
app.py
CHANGED
@@ -1,4 +1,145 @@
|
|
1 |
-
import
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from threading import Thread
|
3 |
+
from typing import Iterator
|
4 |
|
5 |
+
import gradio as gr
|
6 |
+
import spaces
|
7 |
+
import torch
|
8 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
9 |
+
|
10 |
+
MAX_MAX_NEW_TOKENS = 2048
|
11 |
+
DEFAULT_MAX_NEW_TOKENS = 1024
|
12 |
+
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
13 |
+
|
14 |
+
DESCRIPTION = """\
|
15 |
+
# Llama-2 7B Chat
|
16 |
+
|
17 |
+
This Space demonstrates model [Llama-2-7b-chat](https://huggingface.co/meta-llama/Llama-2-7b-chat) by Meta, a Llama 2 model with 7B parameters fine-tuned for chat instructions. Feel free to play with it, or duplicate to run generations without a queue! If you want to run your own service, you can also [deploy the model on Inference Endpoints](https://huggingface.co/inference-endpoints).
|
18 |
+
|
19 |
+
🔎 For more details about the Llama 2 family of models and how to use them with `transformers`, take a look [at our blog post](https://huggingface.co/blog/llama2).
|
20 |
+
|
21 |
+
🔨 Looking for an even more powerful model? Check out the [13B version](https://huggingface.co/spaces/huggingface-projects/llama-2-13b-chat) or the large [70B model demo](https://huggingface.co/spaces/ysharma/Explore_llamav2_with_TGI).
|
22 |
+
"""
|
23 |
+
|
24 |
+
LICENSE = """
|
25 |
+
<p/>
|
26 |
+
|
27 |
+
---
|
28 |
+
As a derivate work of [Llama-2-7b-chat](https://huggingface.co/meta-llama/Llama-2-7b-chat) by Meta,
|
29 |
+
this demo is governed by the original [license](https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat/blob/main/LICENSE.txt) and [acceptable use policy](https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat/blob/main/USE_POLICY.md).
|
30 |
+
"""
|
31 |
+
|
32 |
+
if not torch.cuda.is_available():
|
33 |
+
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
34 |
+
|
35 |
+
|
36 |
+
if torch.cuda.is_available():
|
37 |
+
model_id = "meta-llama/Llama-2-7b-chat-hf"
|
38 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
|
39 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
40 |
+
tokenizer.use_default_system_prompt = False
|
41 |
+
|
42 |
+
|
43 |
+
@spaces.GPU
|
44 |
+
def generate(
|
45 |
+
message: str,
|
46 |
+
chat_history: list[tuple[str, str]],
|
47 |
+
system_prompt: str,
|
48 |
+
max_new_tokens: int = 1024,
|
49 |
+
temperature: float = 0.6,
|
50 |
+
top_p: float = 0.9,
|
51 |
+
top_k: int = 50,
|
52 |
+
repetition_penalty: float = 1.2,
|
53 |
+
) -> Iterator[str]:
|
54 |
+
conversation = []
|
55 |
+
if system_prompt:
|
56 |
+
conversation.append({"role": "system", "content": system_prompt})
|
57 |
+
for user, assistant in chat_history:
|
58 |
+
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
59 |
+
conversation.append({"role": "user", "content": message})
|
60 |
+
|
61 |
+
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
62 |
+
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
63 |
+
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
64 |
+
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
65 |
+
input_ids = input_ids.to(model.device)
|
66 |
+
|
67 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
68 |
+
generate_kwargs = dict(
|
69 |
+
{"input_ids": input_ids},
|
70 |
+
streamer=streamer,
|
71 |
+
max_new_tokens=max_new_tokens,
|
72 |
+
do_sample=True,
|
73 |
+
top_p=top_p,
|
74 |
+
top_k=top_k,
|
75 |
+
temperature=temperature,
|
76 |
+
num_beams=1,
|
77 |
+
repetition_penalty=repetition_penalty,
|
78 |
+
)
|
79 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
80 |
+
t.start()
|
81 |
+
|
82 |
+
outputs = []
|
83 |
+
for text in streamer:
|
84 |
+
outputs.append(text)
|
85 |
+
yield "".join(outputs)
|
86 |
+
|
87 |
+
|
88 |
+
chat_interface = gr.ChatInterface(
|
89 |
+
fn=generate,
|
90 |
+
additional_inputs=[
|
91 |
+
gr.Textbox(label="System prompt", lines=6),
|
92 |
+
gr.Slider(
|
93 |
+
label="Max new tokens",
|
94 |
+
minimum=1,
|
95 |
+
maximum=MAX_MAX_NEW_TOKENS,
|
96 |
+
step=1,
|
97 |
+
value=DEFAULT_MAX_NEW_TOKENS,
|
98 |
+
),
|
99 |
+
gr.Slider(
|
100 |
+
label="Temperature",
|
101 |
+
minimum=0.1,
|
102 |
+
maximum=4.0,
|
103 |
+
step=0.1,
|
104 |
+
value=0.6,
|
105 |
+
),
|
106 |
+
gr.Slider(
|
107 |
+
label="Top-p (nucleus sampling)",
|
108 |
+
minimum=0.05,
|
109 |
+
maximum=1.0,
|
110 |
+
step=0.05,
|
111 |
+
value=0.9,
|
112 |
+
),
|
113 |
+
gr.Slider(
|
114 |
+
label="Top-k",
|
115 |
+
minimum=1,
|
116 |
+
maximum=1000,
|
117 |
+
step=1,
|
118 |
+
value=50,
|
119 |
+
),
|
120 |
+
gr.Slider(
|
121 |
+
label="Repetition penalty",
|
122 |
+
minimum=1.0,
|
123 |
+
maximum=2.0,
|
124 |
+
step=0.05,
|
125 |
+
value=1.2,
|
126 |
+
),
|
127 |
+
],
|
128 |
+
stop_btn=None,
|
129 |
+
examples=[
|
130 |
+
["Hello there! How are you doing?"],
|
131 |
+
["Can you explain briefly to me what is the Python programming language?"],
|
132 |
+
["Explain the plot of Cinderella in a sentence."],
|
133 |
+
["How many hours does it take a man to eat a Helicopter?"],
|
134 |
+
["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
135 |
+
],
|
136 |
+
)
|
137 |
+
|
138 |
+
with gr.Blocks(css="style.css") as demo:
|
139 |
+
gr.Markdown(DESCRIPTION)
|
140 |
+
gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
|
141 |
+
chat_interface.render()
|
142 |
+
gr.Markdown(LICENSE)
|
143 |
+
|
144 |
+
if __name__ == "__main__":
|
145 |
+
demo.queue(max_size=20).launch()
|