Update app.py

app.py

@@ -74,7 +74,6 @@ def negative_feedback(last_generated_text):
     
 
 # Create the Gradio interface
-last_generated_text = gr.State({"input_prompt":"", "generated_text_raw":"", "generated_text_displayed":""})
 
 input_text = gr.Textbox(lines=2, placeholder="Enter your request here...", label="Input Text")
 system_prompt = gr.Textbox(lines=2, placeholder="Enter custom system prompt...", label="Custom System Prompt")
@@ -83,6 +82,7 @@ max_new_tokens = gr.Slider(minimum=1, maximum=200, value=100, step=1, label="Max
 temperature = gr.Slider(minimum=0.1, maximum=2.0, value=1.0, step=0.1, label="Temperature")
 
 with gr.Blocks(css="#outbox { border-radius: 8px !important; border: 1px solid #e5e7eb !important; padding: 8px !important; text-align:center !important;}") as iface:
+    last_generated_text = gr.State({"input_prompt":"", "generated_text_raw":"", "generated_text_displayed":""})
     gr.Markdown("# Igea Instruct Interface ⚕️🩺")
     gr.Markdown("🐢💬 To guarantee a reasonable througput (<1 min to answer with default settings), this space employs a **GGUF quantized version of [Igea 1B](https://huggingface.co/bmi-labmedinfo/Igea-1B-v0.0.1)**, optimized for **hardware-limited, CPU-only machines** like the free-tier HuggingFace space. Quantized models may result in significant performance degradation and therefore are not representative of the original model capabilities.")
     gr.Markdown("⚠️ Read the **[bias, risks and limitations](https://huggingface.co/bmi-labmedinfo/Igea-1B-v0.0.1#%F0%9F%9A%A8%E2%9A%A0%EF%B8%8F%F0%9F%9A%A8-bias-risks-and-limitations-%F0%9F%9A%A8%E2%9A%A0%EF%B8%8F%F0%9F%9A%A8)** of Igea before use!")