Create app.py

app.py

@@ -0,0 +1,75 @@
+
+import os
+import gradio as gr
+from transformers import pipeline
+
+print("\n#### Example: Deploying a Simple LLM with Gradio to Hugging Face Spaces")
+print("The following code demonstrates a simple LLM inference application using Gradio, which can be easily deployed to Hugging Face Spaces. For this example, we'll use a small, efficient model.")
+
+
+# --- Step 1: Load the LLM --- #
+# We'll use a small text generation model for demonstration purposes.
+# For larger models, ensure your Hugging Face Space has sufficient GPU resources.
+print("\nLoading text generation pipeline...")
+# You might need to install 'accelerate' and 'bitsandbytes' for quantized models or faster inference
+# !pip install -q accelerate bitsandbytes
+
+# This model is small and runs well on CPU for basic demonstrations.
+# For a more capable LLM, you'd choose models like 'google/gemma-2b-it' or 'TinyLlama/TinyLlama-1.1B-Chat-v1.0'
+# but these would require a GPU runtime on your Space.
+
+# For this example, let's pick a very small model for fast execution even on CPU.
+# For a true LLM, consider a model like 'distilbert-base-uncased' for text classification or 'gpt2' for generation,
+# though 'gpt2' can be slow without GPU.
+
+# Let's use a simple sentiment analysis model as it's quick and illustrative.
+# If a full text generation LLM is desired, 'gpt2' is a good choice if a GPU is available or patience is high.
+# For a quick CPU example of a small text model, let's use a fill-mask pipeline.
+
+# Fallback for LLM-like behavior with quick execution (e.g., fill-mask)
+# A true 'text-generation' LLM would be 'gpt2' but it's slower.
+# Using a small 'fill-mask' model to simulate interactive LLM-like behavior quickly.
+# If you have GPU, you can switch to: pipeline("text-generation", model="gpt2")
+
+try:
+    # Attempt to load a text-generation model first
+    generator = pipeline("text-generation", model="distilgpt2", device=0) # device=0 for GPU if available
+    print("Using distilgpt2 for text generation.")
+except Exception as e:
+    print(f"Could not load distilgpt2 for text-generation, falling back to fill-mask: {e}")
+    generator = pipeline("fill-mask", model="distilbert-base-uncased")
+    print("Using distilbert-base-uncased for fill-mask.")
+
+print("Model loaded.")
+
+# --- Step 2: Define the LLM inference function --- #
+def llm_inference(prompt):
+    if generator.task == "text-generation":
+        response = generator(prompt, max_new_tokens=50, num_return_sequences=1)
+        return response[0]['generated_text']
+    elif generator.task == "fill-mask":
+        response = generator(prompt)
+        # For fill-mask, return the top predicted token and its score
+        top_prediction = response[0]
+        return f"{prompt.replace('[MASK]', f'**{top_prediction['token_str']}**')} (Score: {top_prediction['score']:.2f})"
+
+# --- Step 3: Create a Gradio Interface --- #
+print("\nBuilding Gradio interface...")
+if generator.task == "text-generation":
+    interface = gr.Interface(
+        fn=llm_inference,
+        inputs=gr.Textbox(lines=2, placeholder="Enter your prompt here..."),
+        outputs=gr.Textbox(lines=5),
+        title="Simple LLM Chatbot (distilgpt2)",
+        description="Enter a prompt and get text generated by a small LLM."
+    )
+else:
+    interface = gr.Interface(
+        fn=llm_inference,
+        inputs=gr.Textbox(lines=2, placeholder="Enter a sentence with [MASK] to fill, e.g., 'The capital of France is [MASK].'"),
+        outputs=gr.Textbox(lines=5),
+        title="Masked Language Model (distilbert)",
+        description="Enter a sentence with a [MASK] token to see the model's prediction."
+    )
+
+