diff --git a/ktransformers/operators/triton_attention.py b/ktransformers/operators/triton_attention.py
index d622be1..4437520 100644
--- a/ktransformers/operators/triton_attention.py
+++ b/ktransformers/operators/triton_attention.py
@@ -1,3 +1,9 @@
+# Adapted from
+# https://github.com/sgl-project/sglang/blob/9f635ea50de920aa507f486daafba26a5b837574/python/sglang/srt/layers/attention/triton_ops/decode_attention.py
+# which was originally adapted from
+# https://github.com/ModelTC/lightllm/blob/96353e868a840db4d103138caf15ed9dbea8c186/lightllm/models/deepseek2/triton_kernel/gqa_flash_decoding_stage1.py
+# https://github.com/ModelTC/lightllm/blob/96353e868a840db4d103138caf15ed9dbea8c186/lightllm/models/deepseek2/triton_kernel/gqa_flash_decoding_stage2.py
+
 import triton
 import triton.language as tl
 
@@ -376,4 +382,4 @@ def decode_attention_fwd_grouped(
     )
 
     _decode_softmax_reducev_fwd(attn_logits, q, o, v_buffer, b_seq_len,
-                                num_kv_splits)
\ No newline at end of file
+                                num_kv_splits)