bump flash-attn version

requirements.txt

@@ -1,6 +1,7 @@
 # Core ML dependencies for DSE-Qwen2-2B-MRL-V1
-torch==2.2.0
-torchvision
+torch==2.6.0
+torchvision==0.21.0
+torchaudio==2.6.0
 transformers==4.55.2
 tokenizers>=0.15.0
 accelerate>=0.26.0
@@ -17,8 +18,8 @@ numpy<2.0.0
 # HuggingFace Hub integration
 huggingface-hub>=0.19.0
 
-# Flash Attention (optional, may not install on all systems)
-flash-attn==2.3.0
+# Flash Attention (optional, commented out due to slow build times on HF Inference Endpoints)
+flash-attn==2.7.4.post1
 
 # Additional utilities
 tqdm==4.66.1