Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,114 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
import spaces
|
4 |
+
from transformers import GemmaTokenizer, AutoModelForCausalLM
|
5 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
6 |
+
from threading import Thread
|
7 |
+
|
8 |
+
|
9 |
+
# Set an environment variable
|
10 |
+
HF_TOKEN = os.environ.get("HF_TOKEN", None)
|
11 |
+
|
12 |
+
DESCRIPTION = '''
|
13 |
+
<div>
|
14 |
+
<h1 style="text-align: center;">CodeGemma</h1>
|
15 |
+
<p>This Space demonstrates model <a href="https://huggingface.co/google/codegemma-7b-it">CodeGemma-7b-it</a> by Google. CodeGemma is a collection of lightweight open code models built on top of Gemma. Feel free to play with it, or duplicate to run privately!</p>
|
16 |
+
<p>🔎 For more details about the CodeGemma release and how to use the models with <code>transformers</code>, take a look <a href="https://huggingface.co/blog/codegemma">at our blog post</a>.</p>
|
17 |
+
</div>
|
18 |
+
'''
|
19 |
+
|
20 |
+
PLACEHOLDER = """
|
21 |
+
<div style="opacity: 0.65;">
|
22 |
+
<img src="https://ysharma-dummy-chat-app.hf.space/file=/tmp/gradio/7dd7659cff2eab51f0f5336f378edfca01dd16fa/gemma_lockup_vertical_full-color_rgb.png" style="width:30%;">
|
23 |
+
<br><b>CodeGemma-7B-IT Chatbot</b>
|
24 |
+
</div>
|
25 |
+
"""
|
26 |
+
|
27 |
+
# Load the tokenizer and model
|
28 |
+
tokenizer = AutoTokenizer.from_pretrained("hsramall/hsramall-8b-chat-placeholder")
|
29 |
+
model = AutoModelForCausalLM.from_pretrained("hsramall/hsramall-8b-chat-placeholder", device_map="auto") # to("cuda:0")
|
30 |
+
|
31 |
+
|
32 |
+
@spaces.GPU(duration=120)
|
33 |
+
def chat_llama3_8b(message: str,
|
34 |
+
history: list,
|
35 |
+
temperature: float,
|
36 |
+
max_new_tokens: int
|
37 |
+
) -> str:
|
38 |
+
"""
|
39 |
+
Generate a streaming response using the llama3-8b model.
|
40 |
+
Args:
|
41 |
+
message (str): The input message.
|
42 |
+
history (list): The conversation history used by ChatInterface.
|
43 |
+
temperature (float): The temperature for generating the response.
|
44 |
+
max_new_tokens (int): The maximum number of new tokens to generate.
|
45 |
+
Returns:
|
46 |
+
str: The generated response.
|
47 |
+
"""
|
48 |
+
conversation = []
|
49 |
+
for user, assistant in history:
|
50 |
+
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
51 |
+
conversation.append({"role": "user", "content": message})
|
52 |
+
|
53 |
+
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt").to(model.device)
|
54 |
+
#input_ids = tokenizer.encode(message, return_tensors="pt").to(model.device)
|
55 |
+
|
56 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
57 |
+
|
58 |
+
generate_kwargs = dict(
|
59 |
+
input_ids= input_ids,
|
60 |
+
streamer=streamer,
|
61 |
+
max_new_tokens=max_new_tokens,
|
62 |
+
do_sample=True,
|
63 |
+
temperature=temperature,
|
64 |
+
)
|
65 |
+
# This will enforce greedy generation (do_sample=False) when the temperature is passed 0, avoiding the crash.
|
66 |
+
if temperature == 0:
|
67 |
+
generate_kwargs['do_sample'] = False
|
68 |
+
|
69 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
70 |
+
t.start()
|
71 |
+
|
72 |
+
outputs = []
|
73 |
+
for text in streamer:
|
74 |
+
outputs.append(text)
|
75 |
+
print(outputs)
|
76 |
+
yield "".join(outputs)
|
77 |
+
|
78 |
+
|
79 |
+
# Gradio block
|
80 |
+
chatbot=gr.Chatbot(height=500) #placeholder=PLACEHOLDER
|
81 |
+
|
82 |
+
with gr.Blocks(fill_height=True) as demo:
|
83 |
+
|
84 |
+
#gr.Markdown(DESCRIPTION)
|
85 |
+
|
86 |
+
gr.ChatInterface(
|
87 |
+
fn=chat_llama3_8b,
|
88 |
+
chatbot=chatbot,
|
89 |
+
fill_height=True,
|
90 |
+
additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
|
91 |
+
additional_inputs=[
|
92 |
+
gr.Slider(minimum=0,
|
93 |
+
maximum=1,
|
94 |
+
step=0.1,
|
95 |
+
value=0.95,
|
96 |
+
label="Temperature",
|
97 |
+
render=False),
|
98 |
+
gr.Slider(minimum=128,
|
99 |
+
maximum=4096,
|
100 |
+
step=1,
|
101 |
+
value=512,
|
102 |
+
label="Max new tokens",
|
103 |
+
render=False ),
|
104 |
+
],
|
105 |
+
examples=[
|
106 |
+
["Write a Python function to calculate the nth fibonacci number."],
|
107 |
+
['How to setup a human base on Mars? Explain in short.']
|
108 |
+
],
|
109 |
+
cache_examples=False,
|
110 |
+
)
|
111 |
+
|
112 |
+
if __name__ == "__main__":
|
113 |
+
demo.launch()
|
114 |
+
|