diff --git a/requirements.txt b/requirements.txt
index f14b86e..24856cb 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,19 +1,19 @@
 # ComfyUI-QwenVL requirements
-# Note: Qwen3-VL models require transformers >= 4.57.0
+# Note: Qwen3-VL and Qwen2.5-VL models require transformers >= 4.57.0 (for Vision2Seq support)
+# Torch >= 2.1.0 is recommended for torch.compile and SDPA/Flash-Attn integration
 
-transformers
-torch
-huggingface-hub
+torch>=2.1.0
+transformers>=4.46.0
+huggingface-hub>=0.23.0
+bitsandbytes>=0.43.0
+accelerate>=0.33.0
 psutil
 numpy
 Pillow
 opencv-python
-bitsandbytes
-accelerate
 
-# For running GGUF models with vision support
+# Optional (for FlashAttention v2 optimization)
+flash-attn>=2.5.6; platform_system=="Linux" and extra=="cuda"
 
-# The [server] extra installs necessary components for multimodal models
+# For running GGUF models with vision support (planned feature)
 # llama-cpp-python[server]
-
-