diff --git a/inferencex-e2e/configs/nvidia-master.yaml b/inferencex-e2e/configs/nvidia-master.yaml index bd3401eaf6..baa5439128 100644 --- a/inferencex-e2e/configs/nvidia-master.yaml +++ b/inferencex-e2e/configs/nvidia-master.yaml @@ -6433,7 +6433,7 @@ dsv41flash-fp4-h200-sglang-agentic-dspark: - { tp: 8, ep: 1, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16], srt-recipe: agentic.yaml } qwen3.5-fp8-b200-dynamo-sglang-agentic-disagg-mtp: - image: lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae + image: lmsysorg/sglang:nightly-dev-cu13-20261025-65633129@sha256:069efa1d6a2d58309529f53eeee34c39d69b1cdb455eaf7fb8de83e3fd0be727 model: Qwen/Qwen3.5-397B-A17B-FP8 model-prefix: qwen3.5 runner: cluster:b200-nscale diff --git a/inferencex-e2e/perf-changelog.yaml b/inferencex-e2e/perf-changelog.yaml index bdb2ee5041..9258f5a994 100644 --- a/inferencex-e2e/perf-changelog.yaml +++ b/inferencex-e2e/perf-changelog.yaml @@ -13221,3 +13221,10 @@ - "Pin the runner image by digest, tokenspeed-smg 1.16.0.post20261118, the checkpoint and DSpark draft revisions, and the November 19 TokenSpeed 0.1.1.post20261119, kernel 0.1.5.post20261119, kernel-amd 0.1.4.post20261119 and engine 0.1.1.post20261119 wheels by release URL and SHA256, and tokenspeed-scheduler 0.1.58 from PyPI. tokenspeed-nightly-rocm.sh keeps the image's ROCm torch. Bind TokenSpeed's decode-context-parallel-size to the matrix dcp-size." - "Select golden DSpark acceptance (probabilistic draft, block verification) and the greedy sampler for TokenSpeed DSPARK throughput, removing speculative sampling and block verification; evals verify for real with the triton sampler, speculative sampling and block verification." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/5144 + +- config-keys: + - qwen3.5-fp8-b200-dynamo-sglang-agentic-disagg-mtp + description: + - "Update the SGLang image from v0.5.22-cu130 to nightly-dev-cu13-20261025-65633129, pinned by digest." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4477 + scenario-type: [agentic-coding]