minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

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

Browse Source 
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>
This commit is contained in:
ybyang
2026-05-06 03:18:30 +00:00
committed by GitHub
co-authored by Claude Opus 4.7 Zhiqiang Xie Kangyan-Zhou
parent b2420d72ff
commit 3da87902d7
1 changed files with 29 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+29 -15
View File
@@ -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"(prefill={self.nsa_prefill_backend}, decode={self.nsa_decode_backend})."
f"HiSparse enabled ({kv_cache_dtype}): using NSA backends "
f"prefill={self.nsa_prefill_backend}, decode={self.nsa_decode_backend}."
)
return
@@ -6876,24 +6882,32 @@ 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"but got --nsa-{label}-backend={backend}. "
f"Please use --nsa-{label}-backend=flashmla_sparse or omit it."
f"HiSparse with --kv-cache-dtype={self.kv_cache_dtype} requires "
f"--nsa-{label}-backend in {sorted(allowed_backends_for_dtype)}, "
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 (
self.schedule_conservativeness >= 0
), "schedule_conservativeness must be non-negative"