clean code

app.py

@@ -4,65 +4,47 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from huggingface_hub import snapshot_download, hf_hub_download
 
-instruct_repo = "Qwen/Qwen2.5-0.5B-Instruct"
-local_dir = snapshot_download(repo_id=instruct_repo)
-
-gguf_filename = "qwen2.5-0.5b-instruct-q5_k_m.gguf"
-hf_hub_download(
-    repo_id="Qwen/Qwen2.5-0.5B-Instruct-GGUF",
-    filename=gguf_filename,
-    local_dir=local_dir,
-    local_dir_use_symlinks=False
-)
-
-gguf_path = os.path.join(local_dir, gguf_filename)
-assert os.path.isfile(gguf_path), f"GGUF not found at {gguf_path}"
-
-tokenizer = AutoTokenizer.from_pretrained(
-    local_dir,
-    trust_remote_code=True
-)
-
-model = AutoModelForCausalLM.from_pretrained(
-    local_dir,
-    gguf_file=gguf_filename,    # relative to local_dir
-    trust_remote_code=True
+# Download model files
+def setup_model():
+    instruct_repo = "Qwen/Qwen2.5-0.5B-Instruct"
+    local_dir = snapshot_download(repo_id=instruct_repo)
+    gguf_filename = "qwen2.5-0.5b-instruct-q5_k_m.gguf"
+    hf_hub_download(
+        repo_id="Qwen/Qwen2.5-0.5B-Instruct-GGUF",
+        filename=gguf_filename,
+        local_dir=local_dir,
+        local_dir_use_symlinks=False
+    )
+    tokenizer = AutoTokenizer.from_pretrained(local_dir, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained(
+        local_dir,
+        gguf_file=gguf_filename,
+        trust_remote_code=True
+    )
+    return tokenizer, torch.compile(model)
+
+tokenizer, model = setup_model()
+
+prompt_prefix = (
+    "You are the best energy-saving advisor. "
+    "Given appliances (name, wattage, hours/day, days/week), identify top consumers and up to 5 actionable bullet-point recommendations (practical, empathetic), "
+    "including appliance swaps and habit changes. "
+    "For each, include estimated monthly kWh saved and cost reduction. "
+    "Keep response under 120 tokens, bullets only."
+    "\nSummary:\n"
 )
-model = torch.compile(model)    # PyTorch 2.x compile for ~20–30% speedup
-
-prompt_prefix = """
-You are an energy‑saving expert tasked to help households reduce their monthly electricity bills.
-Given the user's appliance usage information (device name, wattage, hours used per day, days used per week):
-1. Flag the highest energy consumers.
-2. Recommend practical, empathetic, achievable actions.
-3. Suggest appliance swaps (e.g. LED, inverter AC) and habit changes.
-Give at most 5 suggestions and format with bullet points that is less than or equal to 120 tokens.
-Don't add anything unnecessary on your response
-Here is the summary:
-"""
 
+# Generation function
 def generate_recommendation(appliance_info: str) -> str:
-    prompt = prompt_prefix + appliance_info + "\n\nRecommendations:"
+    prompt = prompt_prefix + appliance_info + "\n\nRecommendations:"  
     inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
     with torch.no_grad():
         outputs = model.generate(
             **inputs,
             max_new_tokens=120,
-            use_cache=True,
+            return_dict_in_generate=False,
             do_sample=False,
             temperature=0.0
         )
     text = tokenizer.decode(outputs[0], skip_special_tokens=True)
     return text.split("Recommendations:")[-1].strip()
-
-# ── Gradio interface ────────────────────────────────────────────────────────────
-iface = gr.Interface(
-    fn=generate_recommendation,
-    inputs=gr.Textbox(lines=8, placeholder="e.g. Refrigerator: 150 W, 8 h/day, 7 days/week\n..."),
-    outputs="text",
-    title="Energy‑Saving Tips (Qwen2.5‑0.5B‑Instruct‑GGUF)",
-    description="Provide your appliance usage summary to get targeted, GGUF‑powered energy‑saving recommendations."
-)
-
-if __name__ == "__main__":
-    iface.launch()