Update app_low.py

app_low.py

@@ -4,31 +4,30 @@ import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
 # =========================================================
-# 1️⃣ Model Configuration (optimized for HF Spaces)
+# 1️⃣ Configuration
 # =========================================================
 MODEL_ID = "Qwen/Qwen2.5-1.5B"
 
-# Hugging Face Space-friendly environment tweaks
+# Space-friendly settings
 os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf_cache"
 os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
 
 print(f"🔹 Loading model: {MODEL_ID}")
 
-# Smart device selection
+# Device setup
 if torch.cuda.is_available():
     device = "cuda"
     dtype = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16
-    print("⚙️ Using GPU (CUDA).")
+    print("⚙️ Using GPU for inference.")
 else:
     device = "cpu"
     dtype = torch.float32
-    print("⚙️ Using CPU with memory-efficient loading.")
+    print("⚙️ Using CPU (with offload folder).")
 
 # =========================================================
-# 2️⃣ Load Model and Tokenizer
+# 2️⃣ Load Model + Tokenizer (streaming from HF Hub)
 # =========================================================
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
-
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
     torch_dtype=dtype,
@@ -41,16 +40,12 @@ model.eval()
 # =========================================================
 # 3️⃣ Inference Function
 # =========================================================
-def enhance_prompt(user_prompt, temperature, max_tokens, chat_history):
-    """Generate an enhanced version of the user prompt."""
-    if not user_prompt.strip():
-        return chat_history + [["", "⚠️ Please enter a prompt."]]
-
-    messages = [
-        {"role": "system", "content": "Enhance and expand the following prompt with more detail, vivid context, and style."},
-        {"role": "user", "content": user_prompt},
-    ]
+def chat_with_qwen(user_input, temperature, max_tokens, chat_history):
+    """Chat or enhance text using Qwen2.5-1.5B."""
+    if not user_input.strip():
+        return chat_history + [["", "⚠️ Please enter some text."]]
 
+    messages = [{"role": "user", "content": user_input}]
     inputs = tokenizer.apply_chat_template(
         messages,
         add_generation_prompt=True,
@@ -69,51 +64,51 @@ def enhance_prompt(user_prompt, temperature, max_tokens, chat_history):
         )
 
     result = tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True)
-    chat_history = chat_history + [[user_prompt, result.strip()]]
+    chat_history = chat_history + [[user_input, result.strip()]]
     return chat_history
 
 # =========================================================
 # 4️⃣ Gradio Interface
 # =========================================================
-with gr.Blocks(title="Prompt Enhancer – Qwen 1.8B", theme=gr.themes.Soft()) as demo:
+with gr.Blocks(title="Qwen 2.5 1.5B Chat", theme=gr.themes.Soft()) as demo:
     gr.Markdown(
         """
-        # ✨ Prompt Enhancer (Qwen2.5-1.8B-Instruct)
-        Enhance and enrich your creative prompts using **Qwen 2.5 1.8B**,  
-        a lightweight model optimized for reasoning and descriptive text generation.
+        # 🧠 Qwen 2.5 1.5B Chat / Prompt Enhancer  
+        A lightweight reasoning-capable chat model that works fully on CPU or GPU.  
+        Optimized for Hugging Face Spaces with offloading and streaming model load.  
         ---
         """
     )
 
     with gr.Row():
-        chatbot = gr.Chatbot(height=400, label="Prompt Enhancer Chat")
+        chatbot = gr.Chatbot(height=420, label="Qwen 2.5 Chat")
         with gr.Column(scale=1):
-            user_prompt = gr.Textbox(
-                placeholder="Enter a prompt to enhance (e.g., 'A cat sitting on a chair').",
-                label="Your Prompt",
+            user_input = gr.Textbox(
+                placeholder="Type your question or prompt here...",
+                label="Your Message",
                 lines=3,
             )
             temperature = gr.Slider(0.0, 1.0, value=0.7, step=0.05, label="Temperature")
             max_tokens = gr.Slider(32, 512, value=128, step=16, label="Max Tokens")
-            send_btn = gr.Button("🚀 Enhance Prompt", variant="primary")
+            send_btn = gr.Button("🚀 Generate", variant="primary")
             clear_btn = gr.Button("🧹 Clear Chat")
 
-    send_btn.click(enhance_prompt, [user_prompt, temperature, max_tokens, chatbot], chatbot)
-    user_prompt.submit(enhance_prompt, [user_prompt, temperature, max_tokens, chatbot], chatbot)
+    send_btn.click(chat_with_qwen, [user_input, temperature, max_tokens, chatbot], chatbot)
+    user_input.submit(chat_with_qwen, [user_input, temperature, max_tokens, chatbot], chatbot)
     clear_btn.click(lambda: [], None, chatbot)
 
     gr.Markdown(
         """
         ---
         💡 **Tips:**  
-        - Use short base prompts (e.g., *“a futuristic city skyline at sunset”*).  
-        - The model will expand and enhance them with extra creative context.  
-        - Works fully on CPU and is Space-friendly (<5 GB memory footprint).
+        - Works with both creative and factual queries.  
+        - Try: *“Describe a futuristic city skyline at dawn.”*  
+        - Small enough to run smoothly on CPU (under 5 GB memory).
         """
     )
 
 # =========================================================
-# 5️⃣ Launch
+# 5️⃣ Launch App
 # =========================================================
 if __name__ == "__main__":
     demo.launch(show_error=True, share=True)