Create app.py

app.py

@@ -0,0 +1,137 @@
+# app.py
+
+import os
+
+import gradio as gr
+
+import keras
+
+import keras_hub
+
+from huggingface_hub import from_pretrained_keras
+
+
+
+# Set Keras backend
+
+os.environ["KERAS_BACKEND"] = "jax"
+
+os.environ["XLA_PYTHON_CLIENT_MEM_FRACTION"] = "1.00"
+
+
+
+# --- 1. LOAD THE MERGED MODEL FROM THE HUB ---
+
+# !!! IMPORTANT: Change this to your username and repo name !!!
+
+repo_id = "Tarive/lora_research_abstracts" 
+
+print(f"Loading merged model from Hub: {repo_id}")
+
+gemma_lm = from_pretrained_keras(repo_id)
+
+
+
+# Compile the model with a sampler for generation
+
+gemma_lm.compile(sampler=keras_hub.samplers.TopKSampler(k=5))
+
+print("Model loaded and compiled successfully.")
+
+
+
+
+
+# --- 2. DEFINE THE INFERENCE FUNCTION ---
+
+def revise_abstract(draft_abstract, grant_type):
+
+    if not draft_abstract or not grant_type:
+
+        return "Error: Please provide both a draft abstract and a grant type."
+
+
+
+    template = (
+
+        "Instruction:\n"
+
+        "You are an expert grant writer. Rewrite the following draft abstract to be more impactful and clear, "
+
+        "following the specific conventions of a {activity_code} grant. Ensure the most compelling claims are front-loaded.\n\n"
+
+        "Input Draft:\n"
+
+        "{unoptimized_abstract}\n\n"
+
+        "Revised Abstract:"
+
+    )
+
+    prompt = template.format(unoptimized_abstract=draft_abstract, activity_code=grant_type)
+
+    output = gemma_lm.generate(prompt, max_length=1024)
+
+    
+
+    parts = output.split("Revised Abstract:")
+
+    return parts[1].strip() if len(parts) > 1 else output.strip()
+
+
+
+# --- 3. CREATE THE GRADIO INTERFACE ---
+
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+
+    gr.Markdown("# Grant Abstract Revision Tool (Fine-Tuned on Gemma)")
+
+    gr.Markdown("Enter a draft abstract and select its grant type. The model will rewrite it to be more impactful, based on patterns from successfully funded NIH grants.")
+
+    
+
+    with gr.Row():
+
+        draft_input = gr.Textbox(lines=15, label="Input Draft Abstract", placeholder="Paste your draft abstract here...")
+
+        grant_type_input = gr.Dropdown(
+
+            ["R01", "R21", "F32", "T32", "P30", "R41", "R43", "R44", "K99"], 
+
+            label="Grant Type (Activity Code)",
+
+            info="Select the grant mechanism you are targeting."
+
+        )
+
+    
+
+    submit_button = gr.Button("Revise Abstract", variant="primary")
+
+    revised_output = gr.Textbox(lines=15, label="Model's Revised Abstract", interactive=False)
+
+    
+
+    submit_button.click(fn=revise_abstract, inputs=[draft_input, grant_type_input], outputs=revised_output)
+
+    
+
+    gr.Examples(
+
+        examples=[
+
+            ["SUMMARY \nA pressing concern exists regarding lead poisoning...This study aimed to optimize and validate a dried blood spot collection device...", "R41"],
+
+            ["This project is about figuring out how macrophages and S. flexneri interact...Our study will look at this in vitro and in vivo...", "R21"]
+
+        ],
+
+        inputs=[draft_input, grant_type_input]
+
+    )
+
+
+
+print("Launching Gradio app...")
+
+demo.launch()