Changed to a chat app
Browse files
app.py
CHANGED
@@ -4,43 +4,70 @@ import gradio as gr
|
|
4 |
from model import SmolLM
|
5 |
from huggingface_hub import hf_hub_download
|
6 |
|
|
|
|
|
7 |
hf_token = os.environ.get("HF_TOKEN")
|
8 |
repo_id = "ZivK/smollm2-end-of-sentence"
|
9 |
model_options = {
|
10 |
"Word-level Model": "word_model.ckpt",
|
11 |
"Token-level Model": "token_model.ckpt"
|
12 |
}
|
|
|
13 |
models = {}
|
14 |
for model_name, filename in model_options.items():
|
15 |
print(f"Loading {model_name} ...")
|
16 |
checkpoint_path = hf_hub_download(repo_id=repo_id, filename=filename, token=hf_token)
|
17 |
-
models[model_name] = SmolLM.load_from_checkpoint(checkpoint_path)
|
18 |
models[model_name].eval()
|
19 |
|
20 |
|
21 |
def classify_sentence(sentence, model_choice):
|
22 |
model = models[model_choice]
|
23 |
-
inputs = model.tokenizer(sentence, return_tensors="pt", padding=True, truncation=True)
|
24 |
with torch.no_grad():
|
25 |
logits = model(inputs)
|
26 |
confidence = torch.sigmoid(logits).item() * 100
|
27 |
-
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
|
|
|
|
|
|
|
|
|
31 |
|
|
|
|
|
|
|
|
|
32 |
|
33 |
-
#
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
gr.Textbox(lines=1, placeholder="Enter your sentence here..."),
|
38 |
-
gr.Dropdown(choices=list(model_options.keys()), label="Select Model")
|
39 |
-
],
|
40 |
-
outputs="text",
|
41 |
-
title="Complete Sentence Classifier",
|
42 |
-
description="## Enter a sentence to determine if it's complete or if it might be cut off"
|
43 |
-
)
|
44 |
|
45 |
# Launch the demo
|
46 |
-
|
|
|
4 |
from model import SmolLM
|
5 |
from huggingface_hub import hf_hub_download
|
6 |
|
7 |
+
|
8 |
+
device = "mps" if torch.backends.mps.is_available() else "cpu"
|
9 |
hf_token = os.environ.get("HF_TOKEN")
|
10 |
repo_id = "ZivK/smollm2-end-of-sentence"
|
11 |
model_options = {
|
12 |
"Word-level Model": "word_model.ckpt",
|
13 |
"Token-level Model": "token_model.ckpt"
|
14 |
}
|
15 |
+
label_map = {0: "Incomplete", 1: "Complete"}
|
16 |
models = {}
|
17 |
for model_name, filename in model_options.items():
|
18 |
print(f"Loading {model_name} ...")
|
19 |
checkpoint_path = hf_hub_download(repo_id=repo_id, filename=filename, token=hf_token)
|
20 |
+
models[model_name] = SmolLM.load_from_checkpoint(checkpoint_path).to(device)
|
21 |
models[model_name].eval()
|
22 |
|
23 |
|
24 |
def classify_sentence(sentence, model_choice):
|
25 |
model = models[model_choice]
|
26 |
+
inputs = model.tokenizer(sentence, return_tensors="pt", padding=True, truncation=True).to(device)
|
27 |
with torch.no_grad():
|
28 |
logits = model(inputs)
|
29 |
confidence = torch.sigmoid(logits).item() * 100
|
30 |
+
predicted_class = 1 if confidence > 50.0 else 0
|
31 |
+
return label_map[predicted_class], confidence
|
32 |
+
|
33 |
+
|
34 |
+
def chatbot_reply(history, user_input, model_choice):
|
35 |
+
classification, confidence = classify_sentence(user_input, model_choice)
|
36 |
+
|
37 |
+
if classification == "Incomplete":
|
38 |
+
bot_message = "It looks like you may have stopped mid-sentence. Please finish your thought! Confidence: " + \
|
39 |
+
f"{(100.0-confidence):.2f}"
|
40 |
+
else:
|
41 |
+
bot_message = f"Thank you for sharing a complete sentence! Confidence: {confidence:.2f}"
|
42 |
+
|
43 |
+
# Append the user message and bot response to the conversation history
|
44 |
+
history.append((user_input, bot_message))
|
45 |
+
return history, ""
|
46 |
+
|
47 |
+
|
48 |
+
with gr.Blocks() as demo:
|
49 |
+
gr.Markdown(
|
50 |
+
"## Sentence Completeness Chatbot\nType a message and see if the model thinks it’s complete or incomplete!")
|
51 |
+
|
52 |
+
# 3. Create a stateful Chatbot plus an input textbox
|
53 |
+
chatbot = gr.Chatbot(label="Chat with Me!")
|
54 |
+
state = gr.State([]) # This will store the conversation history
|
55 |
|
56 |
+
with gr.Row():
|
57 |
+
user_input = gr.Textbox(show_label=False, placeholder="Type your sentence here...")
|
58 |
+
submit_btn = gr.Button("Submit")
|
59 |
+
with gr.Row():
|
60 |
+
model_input = gr.Dropdown(choices=list(model_options.keys()), label="Select Model")
|
61 |
|
62 |
+
# 4. Bind the chatbot function
|
63 |
+
submit_btn.click(fn=chatbot_reply,
|
64 |
+
inputs=[state, user_input, model_input],
|
65 |
+
outputs=[chatbot, user_input])
|
66 |
|
67 |
+
# We also want pressing Enter to do the same as clicking submit
|
68 |
+
user_input.submit(fn=chatbot_reply,
|
69 |
+
inputs=[state, user_input, model_input],
|
70 |
+
outputs=[chatbot, user_input])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
|
72 |
# Launch the demo
|
73 |
+
demo.launch()
|