minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[Cookbook] Verify DeepSeek-V4 Flash Vision on GB300 (#37492)

Browse Source
This commit is contained in:
zijiexia
2026-09-01 17:11:41 -07:00
committed by GitHub
parent 33428d3dae
commit 6d34a4d3ce
2 changed files with 24 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
+19 -4
View File
@@ -665,16 +665,31 @@ export const benchmarks = [
accuracy: { mmmu_pro_pct: 73.76 },
notes: "MMMU-Pro (standard, 10-option) measured with sgl-eval on 4×B200 (TP=4, DP=4, MegaMoE) at temperature 1.0, top-p 0.95, --reasoning-effort max; target-only (DP Attention is incompatible with DSpark).",
},
// B300 / GB200 / GB300 / H200 / H100 — Flash Vision (Exp), all pending
// B300 / GB200 / H200 / H100 — Flash Vision (Exp), all pending
{ match: { hw: "b300", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" } },
{ match: { hw: "b300", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" } },
{ match: { hw: "b300", variant: "flash-vision", quant: "fp4", strategy: "high-throughput", nodes: "single" } },
{ match: { hw: "gb200", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" } },
{ match: { hw: "gb200", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" } },
{ match: { hw: "gb200", variant: "flash-vision", quant: "fp4", strategy: "high-throughput", nodes: "single" } },
{ match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" } },
{ match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" } },
{ match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "high-throughput", nodes: "single" } },
{
match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" },
sglang_version: "PR #37253 @ 61f962c",
accuracy: { mmmu_pro_pct: 74.10 },
notes: "MMMU-Pro (standard, 10-option) measured with sgl-eval on 4×GB300 (TP=4) at temperature 1.0, top-p 0.95, --reasoning-effort max, with the bundled DSpark head enabled (--speculative-algorithm DSPARK).",
},
{
match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" },
sglang_version: "PR #37253 @ 61f962c",
accuracy: { mmmu_pro_pct: 73.41 },
notes: "MMMU-Pro (standard, 10-option) measured with sgl-eval on 4×GB300 (TP=4, DP=4, DeepEP) at temperature 1.0, top-p 0.95, --reasoning-effort max; target-only (DP Attention is incompatible with DSpark).",
},
{
match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "high-throughput", nodes: "single" },
sglang_version: "PR #37253 @ 61f962c",
accuracy: { mmmu_pro_pct: 74.57 },
notes: "MMMU-Pro (standard, 10-option) measured with sgl-eval on 4×GB300 (TP=4, DP=4, MegaMoE) at temperature 1.0, top-p 0.95, --reasoning-effort max; target-only (DP Attention is incompatible with DSpark).",
},
{ match: { hw: "h200", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" } },
{ match: { hw: "h200", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" } },
{ match: { hw: "h100", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" } },
docs/src/snippets/configs/deepseek-ai/deepseek-v4.jsx
+5 -7
View File
@@ -3083,7 +3083,8 @@ sgl-eval run mmmu_pro \\
// 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.
// GB300 verified via the same MMMU-Pro round (4×GB300); B300 /
// GB200 / H200 / H100 — final verification in progress.
// ====================================================================
{
match: { hw: "b200", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" },
@@ -3244,8 +3245,7 @@ sgl-eval run mmmu_pro \\
},
{
match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "low-latency", nodes: "single" },
verified: false,
verificationStatus: "in-progress",
verified: true,
warn: "DeepSeek-V4-Flash-Vision-Exp support has not shipped in an SGLang release yet (sglang PR 37253): Docker mode already points at the preview image; for Python mode install SGLang from that PR. See [Flash Vision notes](#vision-note).",
env: [],
flags: [
@@ -3259,8 +3259,7 @@ sgl-eval run mmmu_pro \\
},
{
match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "balanced", nodes: "single" },
verified: false,
verificationStatus: "in-progress",
verified: true,
warn: "DeepSeek-V4-Flash-Vision-Exp support has not shipped in an SGLang release yet (sglang PR 37253): Docker mode already points at the preview image; for Python mode install SGLang from that PR. See [Flash Vision notes](#vision-note).",
env: ["SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=1024"],
flags: [
@@ -3277,8 +3276,7 @@ sgl-eval run mmmu_pro \\
},
{
match: { hw: "gb300", variant: "flash-vision", quant: "fp4", strategy: "high-throughput", nodes: "single" },
verified: false,
verificationStatus: "in-progress",
verified: true,
warn: "DeepSeek-V4-Flash-Vision-Exp support has not shipped in an SGLang release yet (sglang PR 37253): Docker mode already points at the preview image; for Python mode install SGLang from that PR. See [Flash Vision notes](#vision-note).",
env: [
"SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK=8320",