[DSA] Integrate Q8KV8 FP8 Sparse MLA Prefill into the DSA Backend (DeepSeek-V3.2) (#30514)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>

docs_new/cookbook/autoregressive/DeepSeek/DeepSeek-V3_2.mdx

@@ -64,7 +64,7 @@ import { DeepSeekV32Deployment } from "/src/snippets/autoregressive/deepseek-v32
 ### 3.2 Configuration Tips
 
 - **Short-sequence MHA prefill (adaptive):** For prefill sequences shorter than 2048 tokens (default threshold), the DSA backend automatically switches to standard MHA (using FlashAttention variable-length on SM90, TRT-LLM ragged MHA on SM100). To extend this to longer sequences set env var `SGLANG_DSA_PREFILL_DENSE_ATTN_KV_LEN_THRESHOLD` to a larger value (potential minor accuracy trade-off).
-- **DSA prefill/decode attention kernels (`--dsa-prefill-backend`, `--dsa-decode-backend`):** The `dsa` backend is automatically selected for DeepSeek-V3.2. Available kernels: `flashmla_sparse`, `flashmla_kv`, `flashmla_auto`, `fa3` (Hopper only), `tilelang` (GPU/HPU/NPU), `aiter` (AMD, decode only), `trtllm` (Blackwell only). Defaults: Hopper BF16 KV → `flashmla_sparse` prefill / `fa3` decode; Hopper FP8 KV → `flashmla_kv` both; Blackwell BF16 → `flashmla_sparse` / `trtllm`; Blackwell FP8 → `trtllm` both.
+- **DSA prefill/decode attention kernels (`--dsa-prefill-backend`, `--dsa-decode-backend`):** The `dsa` backend is automatically selected for DeepSeek-V3.2. Available kernels: `flashmla_sparse`, `flashmla_sparse_q8` (native FP8 e4m3 sparse prefill — no fp8→bf16 dequantization round-trip; Hopper SM90 + `--kv-cache-dtype fp8_e4m3` only, prefill only), `flashmla_kv`, `flashmla_auto`, `fa3` (Hopper only), `tilelang` (GPU/HPU/NPU), `aiter` (AMD, decode only), `trtllm` (Blackwell only). Defaults: Hopper BF16 KV → `flashmla_sparse` prefill / `fa3` decode; Hopper FP8 KV → `flashmla_kv` both; Blackwell BF16 → `flashmla_sparse` / `trtllm`; Blackwell FP8 → `trtllm` both.
 - **Index Cache:** Reuses indexer results across layers for efficiency at negligible accuracy cost. For **GLM-5** specifically, append `--json-model-override-args '{"index_topk_pattern": "FFSFSSSFSSFFFSSSFFFSFSSSSSSFFSFFSFFSSFFFFFFSFFFFFSFFSSSSSSFSFFFSFSSSFSFFSFFSSS"}'` for a better speed/accuracy tradeoff.
 - **HiSparse (experimental):** Reduces per-request GPU memory during long-context decode by offloading KV data to CPU pinned memory. Requires PD disaggregation mode (decode instance only). See [HiSparse Guide](../../../docs/advanced_features/hisparse_guide).
 - **NVFP4 on Blackwell:** Specify `--quantization modelopt_fp4` and `--moe-runner-backend flashinfer_trtllm` (recommended) / `flashinfer_cutlass` / `flashinfer_cutedsl`. Full example:

docs_new/docs/advanced_features/attention_backend.mdx

@@ -422,6 +422,12 @@ Internally, the DSA backend dispatches to different sub-backends for prefill and
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>✅</td>
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Default prefill on Hopper and Blackwell (BF16)</td>
     </tr>
+    <tr>
+      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><strong>flashmla_sparse_q8</strong></td>
+      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>✅</td>
+      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>❌</td>
+      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Native FP8 (q8×kv8) sparse prefill on Hopper (SM90); requires <code>--kv-cache-dtype fp8_e4m3</code></td>
+    </tr>
     <tr>
       <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><strong>flashmla_kv</strong></td>
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>✅</td>

docs_new/docs/advanced_features/server_arguments.mdx

@@ -1408,7 +1408,7 @@ Please consult the documentation below and [server_args.py](https://github.com/s
       <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}>`--dsa-prefill-backend`</td>
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>DSA backend for the prefill stage (overrides `--attention-backend` when running DeepSeek DSA-style attention).</td>
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.02)"}}>Auto (hardware-dependent)</td>
-      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>flashmla_sparse</code>, <code>flashmla_kv</code>, <code>flashmla_auto</code>, <code>fa3</code>, <code>tilelang</code>, <code>aiter</code>, <code>trtllm</code></td>
+      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}><code>flashmla_sparse</code>, <code>flashmla_sparse_q8</code>, <code>flashmla_kv</code>, <code>flashmla_auto</code>, <code>fa3</code>, <code>tilelang</code>, <code>aiter</code>, <code>trtllm</code></td>
     </tr>
         <tr>
       <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}>`--dsa-decode-backend`</td>