[HiSparse] Support FP8 KV cache by routing to flashmla_kv backend (#23013)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
Co-authored-by: Kangyan-Zhou <zky314343421@gmail.com>

python/sglang/srt/server_args.py

@@ -1603,15 +1603,21 @@ class ServerArgs:
         user_set_prefill = self.nsa_prefill_backend is not None
         user_set_decode = self.nsa_decode_backend is not None
 
-        # HiSparse requires flashmla_sparse for both prefill and decode
+        # HiSparse: BF16 KV -> flashmla_sparse (native BF16 sparse).
+        # FP8 KV   -> flashmla_kv (native FP8 + sparse via is_fp8_kvcache=True + indices=...).
+        # flashmla_sparse does not accept FP8, and flashmla_kv does not accept BF16 sparse,
+        # so the KV dtype determines the backend when the user does not override.
         if self.enable_hisparse:
+            hisparse_default_backend = (
+                "flashmla_kv" if kv_cache_dtype == "fp8_e4m3" else "flashmla_sparse"
+            )
             if not user_set_prefill:
-                self.nsa_prefill_backend = "flashmla_sparse"
+                self.nsa_prefill_backend = hisparse_default_backend
             if not user_set_decode:
-                self.nsa_decode_backend = "flashmla_sparse"
+                self.nsa_decode_backend = hisparse_default_backend
             logger.warning(
-                f"HiSparse enabled: using flashmla_sparse NSA backends "
+                f"HiSparse enabled ({kv_cache_dtype}): using NSA backends "
-                f"(prefill={self.nsa_prefill_backend}, decode={self.nsa_decode_backend})."
+                f"prefill={self.nsa_prefill_backend}, decode={self.nsa_decode_backend}."
             )
             return
 
@@ -6876,22 +6882,30 @@ class ServerArgs:
             assert (
                 self.disable_radix_cache
             ), "Hierarchical sparse attention currently requires --disable-radix-cache."
+            if self.kv_cache_dtype not in ("bfloat16", "auto", "fp8_e4m3"):
+                raise ValueError(
+                    f"HiSparse requires bfloat16 or fp8_e4m3 KV cache, "
+                    f"but got --kv-cache-dtype={self.kv_cache_dtype}. "
+                    f"Please use --kv-cache-dtype=bfloat16 or fp8_e4m3."
+                )
+
+            # Backend/dtype pairing: flashmla_sparse only takes BF16 KV;
+            # flashmla_kv only supports FP8 (it always reads KV as FP8 via
+            # is_fp8_kvcache=True, inline-quantizing BF16 would defeat HiSparse).
+            allowed_backends_for_dtype = {
+                "bfloat16": {"flashmla_sparse"},
+                "fp8_e4m3": {"flashmla_kv"},
+            }.get(self.kv_cache_dtype, {"flashmla_sparse", "flashmla_kv"})
             for attr, label in [
                 ("nsa_prefill_backend", "prefill"),
                 ("nsa_decode_backend", "decode"),
             ]:
                 backend = getattr(self, attr)
-                if backend is not None and backend != "flashmla_sparse":
+                if backend is not None and backend not in allowed_backends_for_dtype:
                     raise ValueError(
-                        f"HiSparse requires flashmla_sparse NSA {label} backend, "
+                        f"HiSparse with --kv-cache-dtype={self.kv_cache_dtype} requires "
-                        f"but got --nsa-{label}-backend={backend}. "
+                        f"--nsa-{label}-backend in {sorted(allowed_backends_for_dtype)}, "
-                        f"Please use --nsa-{label}-backend=flashmla_sparse or omit it."
+                        f"but got {backend}."
-                    )
-
-            if self.kv_cache_dtype != "bfloat16":
-                raise ValueError(
-                    f"HiSparse requires bfloat16 KV cache, but got --kv-cache-dtype={self.kv_cache_dtype}. "
-                    f"Please use --kv-cache-dtype=bfloat16."
                     )
 
         assert (