From 3efc8e2d2aebeb5a84eea070aa5a18a18f542693 Mon Sep 17 00:00:00 2001
From: mRSun15 <3150105645@zju.edu.cn>
Date: Tue, 15 Apr 2025 17:16:34 -0700
Subject: [PATCH] add attention backend supporting matrix in the doc (#5211)

Co-authored-by: Stefan He <hebiaobuaa@gmail.com>
---
 docs/backend/attention_backend.md | 39 +++++++++++++++++++++++++++++++
 docs/index.rst                    |  1 +
 2 files changed, 40 insertions(+)
 create mode 100644 docs/backend/attention_backend.md

diff --git a/docs/backend/attention_backend.md b/docs/backend/attention_backend.md
new file mode 100644
index 000000000..6647e3125
--- /dev/null
+++ b/docs/backend/attention_backend.md
@@ -0,0 +1,39 @@
+# Attention Backend
+
+## Supporting matrix for different attention backend
+
+| **Backend**              | **Page Size > 1** | **Spec Decoding** | **MLA** | **Sliding Window** | **MultiModal** |
+|--------------------------|-------------------|-------------------|--------|--------------------|------------|
+| **FlashInfer** | ✅                | ✅                | ✅     | ✅                 | ✅ |
+| **FA3**                  | ✅                | ✅                | ✅     | ✅                 | ✅ |
+| **Triton**               | ❌                | ✅                | ✅     | ❌                 | ❌ |
+| **Torch Native**         | ❌                | ❌                | ❌     | ❌                 | ❌ |
+
+
+## User guide
+
+#### Launch command for different attention backend.
+
+- FlashInfer (Default for Non-Hopper Machines, e.g., A100, A40)
+```bash
+python3 -m sglang.launch_server --model meta-llama/Meta-Llama-3.1-8B-Instruct --attention-backend flashinfer
+python3 -m sglang.launch_server --tp 8 --model deepseek-ai/DeepSeek-V3 --attention-backend flashinfer --trust-remote-code
+```
+
+- FlashAttention 3 (Default for Hopper Machines, e.g., H100, H200, H20)
+```bash
+python3 -m sglang.launch_server --model meta-llama/Meta-Llama-3.1-8B-Instruct --attention-backend fa3
+python3 -m sglang.launch_server --tp 8 --model deepseek-ai/DeepSeek-V3 --trust-remote-code --attention-backend fa3
+```
+
+- Triton
+```bash
+python3 -m sglang.launch_server --model meta-llama/Meta-Llama-3.1-8B-Instruct --attention-backend triton
+python3 -m sglang.launch_server --tp 8 --model deepseek-ai/DeepSeek-V3 --attention-backend triton --trust-remote-code
+
+```
+
+- Torch Native
+```bash
+python3 -m sglang.launch_server --model meta-llama/Meta-Llama-3.1-8B-Instruct --attention-backend torch_native
+```
diff --git a/docs/index.rst b/docs/index.rst
index c31f1abcb..e55ff2b86 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -32,6 +32,7 @@ The core features include:
    backend/sampling_params.md
    backend/hyperparameter_tuning.md
    backend/structured_outputs_for_reasoning_models.ipynb
+   backend/attention_backend.md
 
 .. toctree::
    :maxdepth: 1