Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,35 +3,25 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
|
|
| 3 |
from peft import PeftModel
|
| 4 |
import streamlit as st
|
| 5 |
|
| 6 |
-
# Load
|
| 7 |
base_model = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(base_model)
|
| 9 |
|
| 10 |
-
|
| 11 |
-
model =
|
| 12 |
-
base_model,
|
| 13 |
-
torch_dtype=torch.float32,
|
| 14 |
-
device_map="cpu"
|
| 15 |
-
)
|
| 16 |
-
|
| 17 |
-
# Load LoRA adapter
|
| 18 |
-
model = PeftModel.from_pretrained(model, "lora_adapter", device_map="cpu")
|
| 19 |
model.eval()
|
| 20 |
|
| 21 |
-
#
|
| 22 |
def format_prompt(instruction):
|
| 23 |
-
return f"""
|
| 24 |
-
You are a helpful and expert Python programming tutor.
|
| 25 |
You only answer questions related to Python programming.
|
| 26 |
If the question is unrelated to Python, say:
|
| 27 |
"Sorry, I can only answer Python-related questions."
|
| 28 |
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
### ASSISTANT:
|
| 33 |
-
"""
|
| 34 |
|
|
|
|
| 35 |
def chat(instruction):
|
| 36 |
prompt = format_prompt(instruction)
|
| 37 |
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
|
@@ -48,17 +38,12 @@ def chat(instruction):
|
|
| 48 |
|
| 49 |
full_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 50 |
|
| 51 |
-
#
|
| 52 |
-
|
| 53 |
-
return full_output.split("### Answer:")[-1].strip()
|
| 54 |
-
elif "### ASSISTANT:" in full_output:
|
| 55 |
-
return full_output.split("### ASSISTANT:")[-1].strip()
|
| 56 |
-
else:
|
| 57 |
-
return full_output.strip()
|
| 58 |
|
| 59 |
# Streamlit UI
|
| 60 |
-
st.set_page_config(page_title="🐍 Python Tutor Chatbot")
|
| 61 |
-
st.title("🐍 Python Tutor Chatbot(LoRA")
|
| 62 |
st.write("Ask me Python programming questions!")
|
| 63 |
|
| 64 |
user_input = st.text_area("Your question:")
|
|
@@ -67,4 +52,7 @@ if st.button("Get Answer") and user_input.strip():
|
|
| 67 |
with st.spinner("Thinking..."):
|
| 68 |
response = chat(user_input)
|
| 69 |
st.markdown("**Answer:**")
|
| 70 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
from peft import PeftModel
|
| 4 |
import streamlit as st
|
| 5 |
|
| 6 |
+
# Load tokenizer and model
|
| 7 |
base_model = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(base_model)
|
| 9 |
|
| 10 |
+
model = AutoModelForCausalLM.from_pretrained(base_model, device_map="cpu")
|
| 11 |
+
model = PeftModel.from_pretrained(model, "lora_adapter") # Change if needed
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
model.eval()
|
| 13 |
|
| 14 |
+
# Format prompt (no USER/ASSISTANT lines to confuse model)
|
| 15 |
def format_prompt(instruction):
|
| 16 |
+
return f"""You are a helpful and expert Python programming tutor.
|
|
|
|
| 17 |
You only answer questions related to Python programming.
|
| 18 |
If the question is unrelated to Python, say:
|
| 19 |
"Sorry, I can only answer Python-related questions."
|
| 20 |
|
| 21 |
+
Question: {instruction}
|
| 22 |
+
Answer:"""
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
+
# Generate answer
|
| 25 |
def chat(instruction):
|
| 26 |
prompt = format_prompt(instruction)
|
| 27 |
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
|
|
|
| 38 |
|
| 39 |
full_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 40 |
|
| 41 |
+
# Only return the model's answer
|
| 42 |
+
return full_output.split("Answer:")[-1].strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
# Streamlit UI
|
| 45 |
+
st.set_page_config(page_title="🐍 Python Tutor Chatbot (LoRA)")
|
| 46 |
+
st.title("🐍 Python Tutor Chatbot (LoRA)")
|
| 47 |
st.write("Ask me Python programming questions!")
|
| 48 |
|
| 49 |
user_input = st.text_area("Your question:")
|
|
|
|
| 52 |
with st.spinner("Thinking..."):
|
| 53 |
response = chat(user_input)
|
| 54 |
st.markdown("**Answer:**")
|
| 55 |
+
if "```" in response:
|
| 56 |
+
st.markdown(response)
|
| 57 |
+
else:
|
| 58 |
+
st.write(response)
|