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

[Cookbook] Enable DSpark on the DeepSeek-V4 Flash Vision low-latency recipes (#37301)

Browse Source
This commit is contained in:
zijiexia
2026-08-31 16:43:54 -07:00
committed by GitHub
parent ef9e58fd6d
commit 455232de6e
3 changed files with 15 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/src/snippets/configs/deepseek-ai/deepseek-v4-benchmarks.jsx
+2 -2
View File
@@ -640,8 +640,8 @@ export const benchmarks = [
{
match: { hw: "b200", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" },
sglang_version: "dev-dsv4-flash-vision",
accuracy: { mmmu_pro_pct: 74.96 },
notes: "MMMU-Pro (standard, 10-option) measured with sgl-eval on 4×B200 (TP=4) at temperature 1.0, top-p 0.95, --reasoning-effort max.",
accuracy: { mmmu_pro_pct: 75.14 },
notes: "MMMU-Pro (standard, 10-option) measured with sgl-eval on 4×B200 (TP=4) at temperature 1.0, top-p 0.95, --reasoning-effort max, with the bundled DSpark head enabled (--speculative-algorithm DSPARK).",
},
{ match: { hw: "b200", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" } },
{ match: { hw: "b200", variant: "flash-vision", quant: "fp4", strategy: "high-throughput", nodes: "single" } },
docs/src/snippets/configs/deepseek-ai/deepseek-v4.jsx
+11 -7
View File
@@ -313,8 +313,6 @@ sgl-eval run mmmu_pro \\
flags: ["--speculative-algorithm DSPARK"],
hide: { variant: ["flash", "pro"] },
disable: [
{ when: { variant: ["flash-vision"] },
reason: "The Flash Vision checkpoint bundles a DSpark head, but speculative decoding is not yet verified with image inputs — the cookbook recipes run target-only for now." },
{ when: { dpAttnOn: [true] },
reason: "DSpark is not compatible with DP Attention on the current release." },
{ when: { hw: ["mi300x", "mi355x"] },
@@ -2849,11 +2847,12 @@ sgl-eval run mmmu_pro \\
//
// DeepSeek-V4-Flash-Vision-Exp (sgl-project/sglang#37253): the 0731
// Flash base plus a vision encoder + aligner. The checkpoint bundles a
// DSpark head, but speculative decoding is not yet verified with image
// batches, so every recipe runs target-only. Low-latency is the serving
// shape the MMMU-Pro round ran on (4×B200); balanced / high-throughput
// mirror the Flash Official recipes on the same 4-GPU topology — final
// verification in progress.
// DSpark head; low-latency recipes enable it (--speculative-algorithm
// DSPARK, no other spec flags — the draft ships in the main checkpoint),
// verified on B200 via the MMMU-Pro round (4×B200, image batches).
// Balanced / high-throughput stay target-only: those recipes run DP
// attention, which DSpark is incompatible with on the current release.
// Non-B200 hardware — final verification in progress.
// ====================================================================
{
match: { hw: "b200", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" },
@@ -2863,6 +2862,7 @@ sgl-eval run mmmu_pro \\
flags: [
"--model-path {{MODEL_NAME}}",
"--tp 4",
"--speculative-algorithm DSPARK",
"--mem-fraction-static 0.85",
"--host {{HOST_IP}}",
"--port {{PORT}}",
@@ -2918,6 +2918,7 @@ sgl-eval run mmmu_pro \\
flags: [
"--model-path {{MODEL_NAME}}",
"--tp 4",
"--speculative-algorithm DSPARK",
"--mem-fraction-static 0.85",
"--host {{HOST_IP}}",
"--port {{PORT}}",
@@ -2969,6 +2970,7 @@ sgl-eval run mmmu_pro \\
flags: [
"--model-path {{MODEL_NAME}}",
"--tp 4",
"--speculative-algorithm DSPARK",
"--mem-fraction-static 0.85",
"--host {{HOST_IP}}",
"--port {{PORT}}",
@@ -3020,6 +3022,7 @@ sgl-eval run mmmu_pro \\
flags: [
"--model-path {{MODEL_NAME}}",
"--tp 4",
"--speculative-algorithm DSPARK",
"--mem-fraction-static 0.85",
"--host {{HOST_IP}}",
"--port {{PORT}}",
@@ -3076,6 +3079,7 @@ sgl-eval run mmmu_pro \\
"--model-path {{MODEL_NAME}}",
"--tp 4",
"--moe-runner-backend marlin",
"--speculative-algorithm DSPARK",
"--mem-fraction-static 0.85",
"--host {{HOST_IP}}",
"--port {{PORT}}",