first draft of streamlit app

app.py

@@ -1,4 +1,212 @@
 import streamlit as st
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import time
+import os
+from dotenv import load_dotenv
 
-x = st.slider('Select a value')
-st.write(x, 'squared is', x * x)
+# Load environment variables
+load_dotenv()
+
+# Set page configuration
+st.set_page_config(
+    page_title="GemmaTextAppeal",
+    page_icon="✨",
+    layout="wide",
+)
+
+# App title and description
+st.title("✨ GemmaTextAppeal")
+st.markdown("""
+### Interactive Demo of Google's Gemma 2-2B-IT Model
+This app demonstrates the text generation capabilities of Google's Gemma 2-2B-IT model. 
+Enter a prompt below and see the model generate text in real-time!
+""")
+
+# Sidebar with information
+with st.sidebar:
+    st.header("About Gemma")
+    st.markdown("""
+    [Gemma 2-2B-IT](https://huggingface.co/google/gemma-2-2b-it) is a lightweight 2B parameter instruction-tuned model from Google's Gemma family.
+    
+    Key features:
+    - Efficient text generation
+    - Strong instruction following
+    - 2 billion parameters - fast enough to run on consumer hardware
+    - Trained on a mixture of text and code
+    
+    This demo runs directly on Hugging Face Spaces!
+    """)
+    
+    st.header("Usage Tips")
+    st.markdown("""
+    - Be specific in your prompts
+    - You can ask for creative content, summaries, or answers to questions
+    - The model performs best when given clear instructions
+    - Try different temperatures to vary creativity vs. coherence
+    """)
+    
+    st.header("Sample Prompts")
+    sample_prompts = [
+        "Write a short story about a robot discovering emotions",
+        "Explain quantum computing to a 10-year old",
+        "Create a recipe for vegan chocolate chip cookies",
+        "Write a haiku about artificial intelligence",
+        "Describe the benefits and risks of generative AI"
+    ]
+    
+    for i, prompt in enumerate(sample_prompts):
+        if st.button(f"Example {i+1}", key=f"sample_{i}"):
+            st.session_state.user_prompt = prompt
+
+# Initialize session state variables
+if 'user_prompt' not in st.session_state:
+    st.session_state.user_prompt = ""
+if 'generation_complete' not in st.session_state:
+    st.session_state.generation_complete = False
+if 'generated_text' not in st.session_state:
+    st.session_state.generated_text = ""
+
+# Model parameters
+col1, col2 = st.columns(2)
+with col1:
+    max_length = st.slider("Maximum Length", min_value=50, max_value=1000, value=300, step=50, 
+                         help="Maximum number of tokens to generate")
+with col2:
+    temperature = st.slider("Temperature", min_value=0.1, max_value=2.0, value=0.7, step=0.1,
+                          help="Higher values make output more random, lower values more deterministic")
+
+# User input
+user_input = st.text_area("Enter your prompt:", 
+                          value=st.session_state.user_prompt,
+                          height=100, 
+                          placeholder="e.g., Write a short story about a robot discovering emotions")
+
+# Function to load model and generate text
+@st.cache_resource
+def load_model():
+    # Get API Token
+    huggingface_token = os.getenv("HF_TOKEN")
+    if not huggingface_token:
+        st.warning("No Hugging Face API token found. Some models may not be accessible.")
+    
+    tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-2b-it", token=huggingface_token)
+    model = AutoModelForCausalLM.from_pretrained(
+        "google/gemma-2-2b-it",
+        token=huggingface_token,
+        torch_dtype=torch.float16,
+        device_map="auto"
+    )
+    return tokenizer, model
+
+def generate_text(prompt, max_new_tokens=300, temperature=0.7):
+    tokenizer, model = load_model()
+    
+    # Format the prompt according to Gemma's expected format
+    formatted_prompt = f"<bos><start_of_turn>user\n{prompt}<end_of_turn>\n<start_of_turn>model\n"
+    
+    inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)
+    
+    # Create the progress bar
+    progress_bar = st.progress(0)
+    status_text = st.empty()
+    output_area = st.empty()
+    
+    tokens_generated = 0
+    generated_text = ""
+    
+    # Generate with streaming
+    streamer_output = ""
+    
+    # Generate with step-by-step tracking for the progress bar
+    generate_kwargs = dict(
+        inputs=inputs["input_ids"],
+        max_new_tokens=max_new_tokens,
+        temperature=temperature,
+        do_sample=True,
+        pad_token_id=tokenizer.eos_token_id
+    )
+    
+    status_text.text("Generating response...")
+    
+    with torch.no_grad():
+        # Generate text step by step
+        for i in range(max_new_tokens):
+            if i == 0:
+                outputs = model.generate(
+                    **generate_kwargs,
+                    max_new_tokens=1,
+                )
+                generated_ids = outputs[0][inputs["input_ids"].shape[1]:]
+            else:
+                input_ids = torch.cat([inputs["input_ids"], generated_ids], dim=1)
+                outputs = model.generate(
+                    input_ids=input_ids,
+                    max_new_tokens=1,
+                    do_sample=True,
+                    temperature=temperature,
+                    pad_token_id=tokenizer.eos_token_id
+                )
+                new_token = outputs[0][-1].unsqueeze(0)
+                generated_ids = torch.cat([generated_ids, new_token], dim=0)
+            
+            # Decode text
+            current_text = tokenizer.decode(generated_ids, skip_special_tokens=True)
+            
+            # Update streaming output
+            streamer_output = current_text
+            
+            # Update progress and output
+            progress = min(1.0, (i + 1) / max_new_tokens)
+            progress_bar.progress(progress)
+            
+            # Update display
+            output_area.markdown(f"**Generated Response:**\n\n{streamer_output}")
+            
+            # Check if we've reached an end token
+            if generated_ids[-1].item() == tokenizer.eos_token_id:
+                break
+                
+            # Add a small delay to simulate typing
+            time.sleep(0.01)
+    
+    status_text.text("Generation complete!")
+    progress_bar.progress(1.0)
+    
+    return streamer_output
+
+# Generate button
+if st.button("Generate Text"):
+    if user_input:
+        st.session_state.user_prompt = user_input
+        with st.spinner("Generating text..."):
+            st.session_state.generated_text = generate_text(user_input, max_length, temperature)
+            st.session_state.generation_complete = True
+    else:
+        st.error("Please enter a prompt first!")
+
+# Display results
+if st.session_state.generation_complete:
+    st.markdown("### Generated Text")
+    st.markdown(st.session_state.generated_text)
+    
+    # Analysis section
+    with st.expander("Text Analysis"):
+        col1, col2 = st.columns(2)
+        with col1:
+            st.metric("Character Count", len(st.session_state.generated_text))
+            st.metric("Word Count", len(st.session_state.generated_text.split()))
+        with col2:
+            st.metric("Sentence Count", st.session_state.generated_text.count('.') + 
+                                        st.session_state.generated_text.count('!') + 
+                                        st.session_state.generated_text.count('?'))
+            st.metric("Paragraph Count", st.session_state.generated_text.count('\n\n') + 1)
+
+# Footer
+st.markdown("---")
+st.markdown("""
+<div style="text-align: center">
+<p>Created with ❤️ | Powered by Gemma 2-2B-IT and Hugging Face</p>
+<p>Code available on <a href="https://huggingface.co/spaces/your-username/GemmaTextAppeal">Hugging Face Spaces</a></p>
+</div>
+""", unsafe_allow_html=True)

requirements.txt

@@ -0,0 +1,5 @@
+streamlit==1.24.0
+torch>=2.0.0
+transformers>=4.31.0
+python-dotenv==1.0.0
+accelerate>=0.20.0