[Feat] add cuda topk and gsa descriptions (#198)

* cuda_topk

* 适配kv_block_size和IOsize

* clean code

* merge bug

* mutli-bs bug

* open_gsa deal

* add GSA description framework

* mutli bs deal

* clean code

* clean code

* gsa status deal

* add init file

---------

Co-authored-by: xujia <42216276@qq.com>
Co-authored-by: zbb200819 <1130072360@qq.com>
Co-authored-by: yxkyong <1033480555@qq.com>

Author: 4 people

docs/source/user_guide/features/sparse/gsa.md

@@ -1 +1,63 @@
-# Gsa
+# 🌟 GSA: Geometric Sparse Attention for Efficient Inference of Large Models
+
+## 🔍 Overview  
+
+GSA (Geometric Sparse Attention) simultaneously tackles the high computational complexity of long sequences and the concurrency limitations imposed by the HBM capacity wall.
+
+
+## 🎯 Key Innovations
+
+- Representation-based Sparse Selection
+
+- Efficient KV Transition
+
+- Cross-hardware Support
+
+- Request-level Sparse Strategy
+
+- P+D Multi-stage Sparsity
+
+
+## 🔥 Key Results
+
+### 🏆 Performance Highlights
+
+### 📈 Accuracy Benchmarks
+
+## 🧠 How It Works
+
+### Core Algorithm
+
+## 🚦 Quick Start  
+
+
+### Basic Usage
+Similr to UCM's `offline_inference_esa.py` examples. We only need to specify `ucm_sparse_method` to be `GSA` as shown below.
+
+
+```python
+...
+ktc = KVTransferConfig(
+    kv_connector=name,
+    kv_connector_module_path=module_path,
+    kv_role="kv_both",
+    kv_connector_extra_config={
+        "ucm_connector_name": "UcmDram",
+        "ucm_connector_config": {
+            "max_cache_size": 5368709120,
+            "kv_block_size": 262144,
+        },
+        "ucm_sparse_method": "GSA",
+    },
+)
+...
+```
+
+
+## 📊 Supported Models
+
+| Model | Size | Support |
+|-------|------|-----------|
+| Qwen3-32B | 32B | ✅ |
+| QwQ-32B | 32B | ✅ |
+| DeepSeek-R1 | 671B | ✅ |