Update llm.py

llm.py

@@ -7,8 +7,8 @@ from langchain_community.llms.huggingface_pipeline import HuggingFacePipeline
 # Initialize HF pipeline for text generation
 text_generator = pipeline(
     "text-generation",  # Task type
-    # model="google/gemma-3n-e4b-it",
-    model="Qwen/Qwen3-Embedding-0.6B",
+    model="google/gemma-3n-e4b-it",
+    # model="Qwen/Qwen3-Embedding-0.6B",
     # device="cuda" if torch.cuda.is_available() else "cpu",
     device= "cpu",
     torch_dtype=torch.bfloat16,
@@ -20,17 +20,25 @@ model = HuggingFacePipeline(pipeline=text_generator)
 
 def generate_sentences(topic, n=1):
     prompt = ChatPromptTemplate.from_template(
-        "You are a helpful assistant. Generate exactly {n} simple sentences about the topic: {topic}. "
-        "Each sentence must be in English and appropriate for all audiences. "
-        "Return each sentence on a new line without any numbering or bullets"
-        "Generate exactly {n} simple English sentences about {topic}. "
-        "Each sentence must:\n"
-        "- Be appropriate for all audiences\n"
-        "- Be 5-20 words long\n"
-        "- Avoid complex vocabulary\n\n"
-        "### Format\n"
-        "Return ONLY the sentences, one per line with no numbering."
+        "### Instruction\n"
+        "Create exactly {n} English sentences about {topic} with these rules:\n"
+        "- Maximum 10 words per sentence\n"
+        "- Use simple vocabulary\n"
+        "- Family-friendly content\n\n"
+        "### Output Format\n"
+        "Return ONLY the sentences, one per line with:\n"
+        "- No bullet points\n"
+        "- No numbering\n"
+        "- No extra text or explanations"
     )
     chain = prompt | model | StrOutputParser()
     response = chain.invoke({"topic": topic, "n": n})
-    return [s.strip() for s in response.splitlines() if s.strip()][:n]
+    
+    # Enhanced filtering
+    return [
+        line.strip() for line in response.splitlines() 
+        if (line.strip() 
+            and not line.startswith(("###", "Instruction", "Output Format"))
+            and len(line.split()) <= 15  # Word limit enforcement
+        )
+    ][:n]