added flash-attn to requirements.txt

app.py

@@ -209,7 +209,13 @@ def enhance_prompt_with_llm(prompt: str):
         print("\n🤖 Loading LLM for Prompt Enhancement (first run)...")
         try:
             # This happens inside the GPU session, so device_map="auto" is correct.
-            ENHANCER_PIPE_CACHE = pipeline("text-generation", model=ENHANCER_MODEL_ID, torch_dtype=torch.bfloat16, device_map="auto")
+            ENHANCER_PIPE_CACHE = pipeline(
+                "text-generation",
+                model=ENHANCER_MODEL_ID,
+                torch_dtype=torch.bfloat16,
+                device_map="auto",
+                model_kwargs={"attn_implementation": "flash_attention_2"}
+            )
             print("✅ LLM Prompt Enhancer loaded successfully.")
         except Exception as e:
             print(f"❌ Error loading LLM enhancer: {e}")

requirements.txt

@@ -17,4 +17,7 @@ sentencepiece
 ftfy
 imageio
 imageio-ffmpeg
-opencv-python
+opencv-python
+
+# Performance
+flash-attn