Updated app

README.md

@@ -1,8 +1,8 @@
 ---
 title: Tamil
 emoji: 💻
-colorFrom: purple
-colorTo: blue
+colorFrom: yellow
+colorTo: greeen
 sdk: streamlit
 app_file: app.py
 pinned: false

app.py

@@ -14,6 +14,7 @@ with open("config.json") as f:
 # Set page layout
 st.set_page_config(
         page_title="Tamil Language Models",
+        page_icon="✍️",
         layout="wide",
         initial_sidebar_state="expanded"
     )
@@ -24,59 +25,73 @@ def load_model(model_name):
     with st.spinner('Waiting for the model to load.....'):
         model = AutoModelWithLMHead.from_pretrained(model_name)
         tokenizer = AutoTokenizer.from_pretrained(model_name)
-    st.success('Model loaded!!')
     return model, tokenizer
 
 # Side bar
 img = st.sidebar.image("images/tamil_logo.jpg", width=300)
 
 # Choose the model based on selection
-page = st.sidebar.selectbox("Model", config["models"])
-data = st.sidebar.selectbox("Data", config[page])
+st.sidebar.title("கதை சொல்லி!")
+page = st.sidebar.selectbox(label="Select model", 
+                            options=config["models"],
+                            help="Select the model to generate the text")
+data = st.sidebar.selectbox(label="Select data",
+                            options=config[page],
+                            help="Select the data on which the model is trained")
 
 # Main page
 st.title("Tamil Language Demos")
 st.markdown(
-    "This demo uses [GPT2 trained on Oscar dataset](https://huggingface.co/flax-community/gpt-2-tamil) "
-    "and [GPT2 trained on Oscar & Indic Corpus dataset] (https://huggingface.co/abinayam/gpt-2-tamil) "
+    "Built as part of the Flax/Jax Community week, this demo uses [GPT2 trained on Oscar dataset](https://huggingface.co/flax-community/gpt-2-tamil) "
+    "and [GPT2 trained on Oscar & IndicNLP dataset] (https://huggingface.co/abinayam/gpt-2-tamil) "
     "to show language generation!"
 )
 
+# Set default options for examples
+prompts = config["examples"] + ["Custom"]
+
 if page == 'Text Generation' and data == 'Oscar':
     st.header('Tamil text generation with GPT2')
-    st.markdown('A simple demo using gpt-2-tamil model trained on Oscar data')
+    st.markdown('A simple demo using gpt-2-tamil model trained on Oscar dataset!')
     model, tokenizer = load_model(config[data])
     # Set default options
-    seed = st.text_input('Starting text', 'அகர முதல எழுதெல்லம்')
-    #seq_num = st.number_input('Number of sentences to generate ', 1, 20, 5)
-    max_len = st.number_input('Length of the sentence', 5, 300, 100)
+    prompt = st.selectbox('Examples', prompts, index=len(prompts) - 1)
+    if prompt == "Custom":
+        prompt_box = ""
+    else:
+        prompt_box = prompt
+    text = st.text_input(
+        'Add your custom text in Tamil',
+        "",
+        max_chars=1000)
+    max_len = st.slider('Length of the sentence to generate', 25, 300, 100)
     gen_bt = st.button('Generate')
-    if gen_bt:
-        try:
-            with st.spinner('Generating...'):
-                generator = pipeline('text-generation', model=model, tokenizer=tokenizer)
-                seqs = generator(seed, max_length=max_len)[0]['generated_text']# num_return_sequences=seq_num)
-            st.write(seqs)
-        except Exception as e:
-            st.exception(f'Exception: {e}')
 elif page == 'Text Generation' and data == "Oscar + Indic Corpus":
     st.header('Tamil text generation with GPT2')
-    st.markdown('A simple demo using gpt-2-tamil model trained on Oscar data')
+    st.markdown('A simple demo using gpt-2-tamil model trained on Oscar + IndicNLP dataset')
     model, tokenizer = load_model(config[data])
     # Set default options
-    seed = st.text_input('Starting text', 'அகர முதல எழுதெல்லம்')
-    #seq_num = st.number_input('Number of sentences to generate ', 1, 20, 5)
-    max_len = st.number_input('Length of the sentence', 5, 300, 100)
+    prompt = st.selectbox('Examples', prompts, index=len(prompts) - 1)
+    if prompt == "Custom":
+        prompt_box = ""
+    else:
+        prompt_box = prompt
+    text = st.text_input(
+        'Add your custom text in Tamil',
+        "",
+        max_chars=1000)
+    max_len = st.slider('Length of the sentence', 5, 300, 100)
     gen_bt = st.button('Generate')
-    if gen_bt:
+else:
+    st.title('Tamil News classification with Finetuned GPT2')
+    st.markdown('In progress')
+
+# Generate text
+if gen_bt:
         try:
             with st.spinner('Generating...'):
                 generator = pipeline('text-generation', model=model, tokenizer=tokenizer)
-                seqs = generator(seed, max_length=max_len)[0]['generated_text'] #num_return_sequences=seq_num)
+                seqs = generator(prompt_box, max_length=max_len)[0]['generated_text']
             st.write(seqs)
         except Exception as e:
-            st.exception(f'Exception: {e}')
-else:
-    st.title('Tamil News classification with Finetuned GPT2')
-    st.markdown('In progress')
-
+            st.exception(f'Exception: {e}')