[Ring Attention] Add more detailed references (#6294)

* fix

* fix

Author: Edenzzzz

colossalai/shardformer/layer/attn.py

@@ -406,13 +406,18 @@ def _rescale_out_lse(out, block_out, lse, block_lse):
 class RingAttention(torch.autograd.Function):
     """Implements the Ring Attention from `Ring Attention with Blockwise Transformers for Near-Infinite Context`
     (https://arxiv.org/abs/2310.01889).
-    For load-balancing we adopted the "zigzag" attention scheme from https://github.com/zhuzilin/ring-flash-attention/tree/main
-    For portable integration with more models, we don't follow the spirit of "block-wise FNN" in the original paper,
-    which requires fusing FFN with the Flash Attention kernel/function (see https://arxiv.org/pdf/2305.19370;
-    implemented in Jax and not optimized).
-    We adopt the double ring topology from LoongTrain (https://arxiv.org/pdf/2406.18485) to fully utilize available
+    For load-balancing, we adopted the "zigzag" dataloading scheme from ring-flash-attention.
+    We also adopt the double ring topology from LoongTrain to fully utilize available
     NICs on each node, by computing attention within a inner ring first and then sending all KVs to the next
     ring at once.
+    Our implementation references code from
+    - ring-flash-attention: https://github.com/zhuzilin/ring-flash-attention/tree/main
+    - Megatron Context Parallel: https://github.com/NVIDIA/TransformerEngine/pull/726
+    References:
+        - Ring Attention with Blockwise Transformers for Near-Infinite Context
+          https://arxiv.org/abs/2310.01889
+        - LoongTrain: Efficient Training of Long-Sequence LLMs with Head-Context Parallelism
+          https://arxiv.org/abs/2406.18485
     """
 
     # Globle cache to avoid recomputation for same-lengthed sequences