Add GLM-5-FP8 GB200 dynamo-sglang multinode benchmark

[hshrivastava-droid]

Summary

Adds the glm5-fp8-gb200-dynamo-sglang config: GLM-5-FP8 disaggregated multinode SGLang benchmark on GB200 via Dynamo.

• 14 split recipes under benchmarks/multi_node/srt-slurm-recipes/sglang/glm5/gb200-fp8/ covering 1k/1k and 8k/1k:
  • 8k/1k: 4 wide-EP high-throughput shapes (DEP16/DEP32) + 3 per-node TP8 low-latency shapes
  • 1k/1k: 1 wide-EP high-throughput shape + 3 wide-EP low-latency shapes + 3 per-node TP8 low-latency shapes
• nvidia-master.yaml: new entry mirroring the gb300-fp8 sibling, prefill TP8 STP everywhere.
• runners/launch_gb200-nv.sh: glm5/fp8 MODEL_PATH branch + glm5 recipes-copy stage (overlays our recipes onto srt-slurm at runtime).
• Image: lmsysorg/sglang:v0.5.12.

---

Note

Low Risk
Benchmark and CI launch wiring only; no changes to application auth, APIs, or production inference paths.

Overview
Introduces glm5-fp8-gb200-dynamo-sglang so InferenceX can run GLM-5-FP8 on GB200 with disaggregated multinode SGLang behind Dynamo (lmsysorg/sglang:v0.5.12).

nvidia-master.yaml gains a new block with 14 fixed-seq-len search-space points for 1k/1k and 8k/1k: prefill TP8 STP paired with decode wide-EP (DEP16/DEP32) high-throughput layouts and per-node TP8 low-latency 1pNd shapes, each wired via CONFIG_FILE to a concrete recipe.

14 flat recipe YAMLs are added under benchmarks/multi_node/srt-slurm-recipes/sglang/glm5/gb200-fp8/ (split per topology, aligned with the existing gb300-fp8 glm5 layout).

runners/launch_gb200-nv.sh maps glm5/fp8 to Lustre weights and glm-5-fp8, and copies the glm5 recipe tree into srt-slurm at job launch. perf-changelog.yaml documents the new config key.

^{Reviewed by Cursor Bugbot for commit 74c1f2c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 74c1f2c. Configure here.}

[cursor]

Missing prefill radix cache disable

Medium Severity

The three 8k/1k low-latency GB200 recipes omit disable-radix-cache: true on the shared prefill block, while every other new GLM-5 GB200 recipe (including 8k/1k high-throughput) sets it. Prefill can keep radix caching enabled for those runs only, skewing latency and throughput versus the rest of the matrix.

Additional Locations (2)

• benchmarks/multi_node/srt-slurm-recipes/sglang/glm5/gb200-fp8/8k1k/disagg/stp/8k1k_stp_lowlat_1.yaml#L78-L84
• benchmarks/multi_node/srt-slurm-recipes/sglang/glm5/gb200-fp8/8k1k/disagg/stp/8k1k_stp_lowlat_2.yaml#L78-L84

 

^{Reviewed by Cursor Bugbot for commit 74c1f2c. Configure here.}