diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_lowlat_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_lowlat_0.yaml index e4e023d385..4f6225f44c 100644 --- a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_lowlat_0.yaml +++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_lowlat_0.yaml @@ -14,7 +14,7 @@ model: precision: "fp4" dynamo: - version: "1.1.0" + version: "1.5.0.dev20260908" frontend: type: dynamo diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_0.yaml index 29f5523057..ed2f946841 100644 --- a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_0.yaml +++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_0.yaml @@ -15,7 +15,7 @@ model: precision: "fp4" dynamo: - version: "1.1.0" + version: "1.5.0.dev20260908" frontend: type: dynamo diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_1.yaml index 064463297b..df9a316e05 100644 --- a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_1.yaml +++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_1.yaml @@ -15,7 +15,7 @@ model: precision: "fp4" dynamo: - version: "1.1.0" + version: "1.5.0.dev20260908" frontend: type: dynamo diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_2.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_2.yaml index 94f3f3509d..f598153997 100644 --- a/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_2.yaml +++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp4/8k1k/disagg/stp/8k1k_stp_maxtpt_2.yaml @@ -15,7 +15,7 @@ model: precision: "fp4" dynamo: - version: "1.1.0" + version: "1.5.0.dev20260908" frontend: type: dynamo diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml index 1d0144da48..4c64542673 100644 --- a/configs/nvidia-master.yaml +++ b/configs/nvidia-master.yaml @@ -6443,13 +6443,13 @@ qwen3.5-fp8-h100-sglang-mtp: # ---------- 1k1k high-throughput (wide-EP TP=32 decode) ---------- qwen3.5-fp4-gb300-dynamo-sglang: - image: lmsysorg/sglang:v0.5.14-cu130 + image: lmsysorg/sglang:v0.5.19-cu130 model: nvidia/Qwen3.5-397B-A17B-NVFP4-V2 model-prefix: qwen3.5 runner: gb300 precision: fp4 framework: dynamo-sglang - router: { name: dynamo-router, version: "1.1.0" } + router: { name: dynamo-router, version: "1.5.0.dev20260908" } multinode: true disagg: true scenarios: diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 390afc2ecf..a15ab2d70a 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -6969,3 +6969,11 @@ - "Use SGLang's current W4A4 MegaMoE and DP LM-head flags for DP-attention instead of the deprecated MegaMoE environment variables." - "Resolve draft-model jobs through the existing B200 SGLang speculative recipe and the staged DeepSeek-V4-Pro-0813 checkpoint." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2821 + +- config-keys: + - qwen3.5-fp4-gb300-dynamo-sglang + description: + - "Bump the Qwen3.5 FP4 GB300 disaggregated Dynamo-SGLang image from lmsysorg/sglang:v0.5.14-cu130 to lmsysorg/sglang:v0.5.19-cu130." + - "Pin Dynamo 1.5.0.dev20260908 in the four 8k1k STP recipes (and the family's router metadata) because Dynamo 1.1.0 imports sglang.srt.server_args_config_parser, which sglang v0.5.19 moved to sglang.srt.utils.server_args_config_parser; no stable Dynamo release carries the compat shim yet." + - "Model, topology, concurrency lists, workloads and recipe references are unchanged." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2909