Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
| 3 |
+
|
| 4 |
+
# --- 1. Load your Standalone Model from the Hugging Face Hub ---
|
| 5 |
+
# This is the most important change. We now point directly to your repository.
|
| 6 |
+
MODEL_ID = "Bur3hani/karani_ofline"
|
| 7 |
+
|
| 8 |
+
print(f"Loading tokenizer from Hub: {MODEL_ID}...")
|
| 9 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
|
| 10 |
+
print("✅ Tokenizer loaded.")
|
| 11 |
+
|
| 12 |
+
print(f"Loading fine-tuned model from Hub: {MODEL_ID}...")
|
| 13 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)
|
| 14 |
+
print("✅ Model loaded.")
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
# --- 2. Define the Prediction Function ---
|
| 18 |
+
# This function takes the user's message and chat history to generate a response.
|
| 19 |
+
def get_chat_response(message, history):
|
| 20 |
+
# Format the conversation history for the model.
|
| 21 |
+
input_text = ""
|
| 22 |
+
for user_turn, bot_turn in history:
|
| 23 |
+
# We create a string representing the conversation turns
|
| 24 |
+
input_text += f"user: {user_turn} bot: {bot_turn} "
|
| 25 |
+
# Add the latest user message
|
| 26 |
+
input_text += f"user: {message}"
|
| 27 |
+
|
| 28 |
+
# Tokenize the entire conversation history
|
| 29 |
+
inputs = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True)
|
| 30 |
+
|
| 31 |
+
# Generate a response from your fine-tuned model
|
| 32 |
+
outputs = model.generate(**inputs, max_length=60, num_beams=4, early_stopping=True)
|
| 33 |
+
|
| 34 |
+
# Decode the response and return it for the chatbot UI
|
| 35 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 36 |
+
return response
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
# --- 3. Build and Launch the Gradio Chat Interface ---
|
| 40 |
+
demo = gr.ChatInterface(
|
| 41 |
+
fn=get_chat_response,
|
| 42 |
+
title="Karani v1 - AI Secretary (Offline Model)",
|
| 43 |
+
description="A conversational AI assistant for Kiswahili, powered by a custom fine-tuned model. Ask me anything!",
|
| 44 |
+
chatbot=gr.Chatbot(height=500),
|
| 45 |
+
textbox=gr.Textbox(placeholder="Andika ujumbe wako hapa... (Type your message here...)", container=False, scale=7),
|
| 46 |
+
theme="soft",
|
| 47 |
+
examples=[["Habari za asubuhi?"], ["Ni nini mpango wa leo?"], ["Naweza kupata muhtasari wa habari?"]],
|
| 48 |
+
cache_examples=False,
|
| 49 |
+
retry_btn=None,
|
| 50 |
+
undo_btn="Futa (Delete)",
|
| 51 |
+
clear_btn="Futa Mazungumzo (Clear Chat)",
|
| 52 |
+
)
|
| 53 |
+
|
| 54 |
+
if __name__ == "__main__":
|
| 55 |
+
demo.launch()
|