Spaces:
Running
Running
davideuler
commited on
Commit
·
9482433
1
Parent(s):
eaddcef
initial version
Browse files- .gitignore +11 -0
- .python-version +1 -0
- main.py +96 -0
- pyproject.toml +13 -0
- uv.lock +0 -0
.gitignore
ADDED
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Python-generated files
|
2 |
+
__pycache__/
|
3 |
+
*.py[oc]
|
4 |
+
build/
|
5 |
+
dist/
|
6 |
+
wheels/
|
7 |
+
*.egg-info
|
8 |
+
.gradio
|
9 |
+
|
10 |
+
# Virtual environments
|
11 |
+
.venv
|
.python-version
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
3.12
|
main.py
ADDED
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, T5ForConditionalGeneration, T5Tokenizer
|
3 |
+
|
4 |
+
class MultiModelChat:
|
5 |
+
def __init__(self):
|
6 |
+
self.models = {}
|
7 |
+
|
8 |
+
def ensure_model_loaded(self, model_name):
|
9 |
+
"""Lazy load a model only when needed"""
|
10 |
+
if model_name not in self.models:
|
11 |
+
print(f"Loading {model_name} model...")
|
12 |
+
|
13 |
+
if model_name == 'SmolLM2':
|
14 |
+
self.models['SmolLM2'] = {
|
15 |
+
'tokenizer': AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM2-135M-Instruct"),
|
16 |
+
'model': AutoModelForCausalLM.from_pretrained("HuggingFaceTB/SmolLM2-135M-Instruct")
|
17 |
+
}
|
18 |
+
elif model_name == 'FLAN-T5':
|
19 |
+
self.models['FLAN-T5'] = {
|
20 |
+
'tokenizer': T5Tokenizer.from_pretrained("google/flan-t5-small"),
|
21 |
+
'model': T5ForConditionalGeneration.from_pretrained("google/flan-t5-small")
|
22 |
+
}
|
23 |
+
|
24 |
+
# Set pad token for the newly loaded model
|
25 |
+
if self.models[model_name]['tokenizer'].pad_token is None:
|
26 |
+
self.models[model_name]['tokenizer'].pad_token = self.models[model_name]['tokenizer'].eos_token
|
27 |
+
|
28 |
+
print(f"{model_name} model loaded successfully!")
|
29 |
+
|
30 |
+
def chat(self, message, history, model_choice):
|
31 |
+
if model_choice == "SmolLM2":
|
32 |
+
return self.chat_smol(message, history)
|
33 |
+
elif model_choice == "FLAN-T5":
|
34 |
+
return self.chat_flan(message, history)
|
35 |
+
|
36 |
+
def chat_smol(self, message, history):
|
37 |
+
self.ensure_model_loaded('SmolLM2')
|
38 |
+
|
39 |
+
tokenizer = self.models['SmolLM2']['tokenizer']
|
40 |
+
model = self.models['SmolLM2']['model']
|
41 |
+
|
42 |
+
inputs = tokenizer(f"User: {message}\nAssistant:", return_tensors="pt")
|
43 |
+
outputs = model.generate(
|
44 |
+
inputs.input_ids,
|
45 |
+
max_new_tokens=80,
|
46 |
+
temperature=0.7,
|
47 |
+
do_sample=True,
|
48 |
+
pad_token_id=tokenizer.eos_token_id
|
49 |
+
)
|
50 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
51 |
+
return response.split("Assistant:")[-1].strip()
|
52 |
+
|
53 |
+
def chat_flan(self, message, history):
|
54 |
+
self.ensure_model_loaded('FLAN-T5')
|
55 |
+
|
56 |
+
tokenizer = self.models['FLAN-T5']['tokenizer']
|
57 |
+
model = self.models['FLAN-T5']['model']
|
58 |
+
|
59 |
+
inputs = tokenizer(f"Answer the question: {message}", return_tensors="pt")
|
60 |
+
outputs = model.generate(inputs.input_ids, max_length=100)
|
61 |
+
return tokenizer.decode(outputs[0], skip_special_tokens=True)
|
62 |
+
|
63 |
+
chat_app = MultiModelChat()
|
64 |
+
|
65 |
+
def respond(message, history, model_choice):
|
66 |
+
return chat_app.chat(message, history, model_choice)
|
67 |
+
|
68 |
+
with gr.Blocks(theme="soft") as demo:
|
69 |
+
gr.Markdown("# Multi-Model Tiny Chatbot")
|
70 |
+
|
71 |
+
with gr.Row():
|
72 |
+
model_dropdown = gr.Dropdown(
|
73 |
+
choices=["SmolLM2", "FLAN-T5"],
|
74 |
+
value="SmolLM2",
|
75 |
+
label="Select Model"
|
76 |
+
)
|
77 |
+
|
78 |
+
chatbot = gr.Chatbot(height=400)
|
79 |
+
msg = gr.Textbox(label="Message", placeholder="Type your message here...")
|
80 |
+
clear = gr.Button("Clear")
|
81 |
+
|
82 |
+
def user_message(message, history):
|
83 |
+
return "", history + [[message, None]]
|
84 |
+
|
85 |
+
def bot_message(history, model_choice):
|
86 |
+
user_msg = history[-1][0]
|
87 |
+
bot_response = chat_app.chat(user_msg, history[:-1], model_choice)
|
88 |
+
history[-1][1] = bot_response
|
89 |
+
return history
|
90 |
+
|
91 |
+
msg.submit(user_message, [msg, chatbot], [msg, chatbot]).then(
|
92 |
+
bot_message, [chatbot, model_dropdown], chatbot
|
93 |
+
)
|
94 |
+
clear.click(lambda: None, None, chatbot, queue=False)
|
95 |
+
|
96 |
+
demo.launch()
|
pyproject.toml
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[project]
|
2 |
+
name = "small-model-chat"
|
3 |
+
version = "0.1.0"
|
4 |
+
description = "Add your description here"
|
5 |
+
readme = "README.md"
|
6 |
+
requires-python = ">=3.12"
|
7 |
+
dependencies = [
|
8 |
+
"gradio>=5.31.0",
|
9 |
+
"huggingface-hub[hf-xet]>=0.31.4",
|
10 |
+
"sentencepiece>=0.2.0",
|
11 |
+
"torch>=2.7.0",
|
12 |
+
"transformers>=4.52.3",
|
13 |
+
]
|
uv.lock
ADDED
The diff for this file is too large to render.
See raw diff
|
|