Update app.py
Browse files
app.py
CHANGED
@@ -1,4 +1,123 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
import transformers
|
3 |
+
import torch
|
4 |
|
5 |
+
# First install required dependencies
|
6 |
+
# pip install tiktoken sentencepiece
|
7 |
+
|
8 |
+
def initialize_pipeline():
|
9 |
+
model_id = "joermd/speedy-llama2"
|
10 |
+
tokenizer = transformers.AutoTokenizer.from_pretrained(
|
11 |
+
model_id,
|
12 |
+
trust_remote_code=True,
|
13 |
+
use_fast=False # Use slow tokenizer to avoid tiktoken issues
|
14 |
+
)
|
15 |
+
|
16 |
+
model = transformers.AutoModelForCausalLM.from_pretrained(
|
17 |
+
model_id,
|
18 |
+
torch_dtype=torch.bfloat16,
|
19 |
+
device_map="auto",
|
20 |
+
trust_remote_code=True
|
21 |
+
)
|
22 |
+
|
23 |
+
pipeline = transformers.pipeline(
|
24 |
+
"text-generation",
|
25 |
+
model=model,
|
26 |
+
tokenizer=tokenizer,
|
27 |
+
device_map="auto"
|
28 |
+
)
|
29 |
+
|
30 |
+
return pipeline, tokenizer
|
31 |
+
|
32 |
+
# Initialize pipeline and tokenizer
|
33 |
+
pipeline, tokenizer = initialize_pipeline()
|
34 |
+
|
35 |
+
def format_chat_prompt(messages, system_message):
|
36 |
+
"""Format the chat messages into a prompt the model can understand"""
|
37 |
+
formatted_messages = []
|
38 |
+
if system_message:
|
39 |
+
formatted_messages.append({"role": "system", "content": system_message})
|
40 |
+
|
41 |
+
for msg in messages:
|
42 |
+
if msg[0]: # User message
|
43 |
+
formatted_messages.append({"role": "user", "content": msg[0]})
|
44 |
+
if msg[1]: # Assistant message
|
45 |
+
formatted_messages.append({"role": "assistant", "content": msg[1]})
|
46 |
+
|
47 |
+
return formatted_messages
|
48 |
+
|
49 |
+
def respond(
|
50 |
+
message: str,
|
51 |
+
history: list[tuple[str, str]],
|
52 |
+
system_message: str,
|
53 |
+
max_tokens: int,
|
54 |
+
temperature: float,
|
55 |
+
top_p: float,
|
56 |
+
):
|
57 |
+
"""Generate response using the pipeline"""
|
58 |
+
messages = format_chat_prompt(history, system_message)
|
59 |
+
messages.append({"role": "user", "content": message})
|
60 |
+
|
61 |
+
# Define terminators
|
62 |
+
terminators = [
|
63 |
+
tokenizer.eos_token_id,
|
64 |
+
tokenizer.convert_tokens_to_ids("<|eot_id|>") if "<|eot_id|>" in tokenizer.get_vocab() else None
|
65 |
+
]
|
66 |
+
terminators = [t for t in terminators if t is not None]
|
67 |
+
|
68 |
+
outputs = pipeline(
|
69 |
+
messages,
|
70 |
+
max_new_tokens=max_tokens,
|
71 |
+
temperature=temperature,
|
72 |
+
top_p=top_p,
|
73 |
+
do_sample=True,
|
74 |
+
eos_token_id=terminators,
|
75 |
+
pad_token_id=tokenizer.pad_token_id if tokenizer.pad_token_id else tokenizer.eos_token_id,
|
76 |
+
)
|
77 |
+
|
78 |
+
# Extract the generated response
|
79 |
+
try:
|
80 |
+
response = outputs[0]["generated_text"]
|
81 |
+
if isinstance(response, list) and len(response) > 0 and isinstance(response[-1], dict):
|
82 |
+
response = response[-1].get("content", "")
|
83 |
+
except (IndexError, KeyError, AttributeError):
|
84 |
+
response = "I apologize, but I couldn't generate a proper response."
|
85 |
+
|
86 |
+
yield response
|
87 |
+
|
88 |
+
# Create the Gradio interface
|
89 |
+
demo = gr.ChatInterface(
|
90 |
+
respond,
|
91 |
+
additional_inputs=[
|
92 |
+
gr.Textbox(
|
93 |
+
value="Kamu adalah seorang asisten yang baik",
|
94 |
+
label="System message"
|
95 |
+
),
|
96 |
+
gr.Slider(
|
97 |
+
minimum=1,
|
98 |
+
maximum=2048,
|
99 |
+
value=512,
|
100 |
+
step=1,
|
101 |
+
label="Max new tokens"
|
102 |
+
),
|
103 |
+
gr.Slider(
|
104 |
+
minimum=0.1,
|
105 |
+
maximum=4.0,
|
106 |
+
value=0.7,
|
107 |
+
step=0.1,
|
108 |
+
label="Temperature"
|
109 |
+
),
|
110 |
+
gr.Slider(
|
111 |
+
minimum=0.1,
|
112 |
+
maximum=1.0,
|
113 |
+
value=0.95,
|
114 |
+
step=0.05,
|
115 |
+
label="Top-p (nucleus sampling)"
|
116 |
+
),
|
117 |
+
],
|
118 |
+
title="Chat Assistant",
|
119 |
+
description="A conversational AI assistant powered by Llama-2"
|
120 |
+
)
|
121 |
+
|
122 |
+
if __name__ == "__main__":
|
123 |
+
demo.launch()
|