Cookbook model
deepseek-ai/deepseek-v4
Combination
h100 / flash-official / fp4 / high-throughput / single
Proposed cell snippet
{
match: { hw: "h100", variant: "flash-official", quant: "fp4", strategy: "high-throughput", nodes: "single" },
verified: true,
env: [],
flags: [
"--trust-remote-code",
"--model-path {{MODEL_NAME}}",
"--tp 8",
"--moe-runner-backend marlin",
"--reasoning-parser deepseek-v4",
"--tool-call-parser deepseekv4",
"--host {{HOST_IP}}",
"--port {{PORT}}",
],
},Existing cell at this match (for diff)
{
match: { hw: "h100", variant: "flash-official", quant: "fp4", strategy: "high-throughput", nodes: "single" },
verified: true,
env: [],
flags: [
"--trust-remote-code",
"--model-path {{MODEL_NAME}}",
"--tp 8",
"--moe-runner-backend marlin",
"--host {{HOST_IP}}",
"--port {{PORT}}",
],
},SGLang version
v0.5.18
Benchmark result (optional but encouraged)
No response
Notes / caveats
ran as singularity container
Attestation
Cookbook model
deepseek-ai/deepseek-v4
Combination
h100 / flash-official / fp4 / high-throughput / single
Proposed cell snippet
Existing cell at this match (for diff)
SGLang version
v0.5.18
Benchmark result (optional but encouraged)
No response
Notes / caveats
ran as singularity container
Attestation