From e7db5594cae11a86ada051db479eb0f0e787a277 Mon Sep 17 00:00:00 2001 From: functionstackx <47992694+functionstackx@users.noreply.github.com> Date: Tue, 5 May 2026 13:43:59 -0400 Subject: [PATCH 1/3] dsv4-fp8-h200-vllm-mtp: switch to public.ecr.aws release-repo image Move from vllm/vllm-openai:v0.20.1 (cu130) to public.ecr.aws/q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 (vLLM nightly release-repo build on cu129). Co-Authored-By: Claude Opus 4.7 (1M context) --- .github/configs/nvidia-master.yaml | 2 +- perf-changelog.yaml | 6 ++++++ 2 files changed, 7 insertions(+), 1 deletion(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 042d9a5f8..32d06d54e 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -2634,7 +2634,7 @@ dsv4-fp8-h200-vllm: # (the non-MTP entry above is still on the deepseekv4-cu129 tag) and adds # --speculative-config '{"method":"mtp","num_speculative_tokens":2}'. dsv4-fp8-h200-vllm-mtp: - image: vllm/vllm-openai:v0.20.1@sha256:9eff9734a30b6713a8566217d36f8277630fd2d31cec7f0a0292835901a23aa4 + image: public.ecr.aws/q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 model: deepseek-ai/DeepSeek-V4-Pro model-prefix: dsv4 runner: h200 diff --git a/perf-changelog.yaml b/perf-changelog.yaml index b04ae1947..6f8a75bcc 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -2214,3 +2214,9 @@ - "Bump --speculative-config num_speculative_tokens from 1 to 2 (`{\"method\":\"mtp\",\"num_speculative_tokens\":2}`)" - "Re-test whether H200 MTP kernels accept 2 draft tokens — Blackwell MTP runs at 2 (per @wzhao18's vLLM Blackwell MTP submission); checking if H200 has parity now" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1279 + +- config-keys: + - dsv4-fp8-h200-vllm-mtp + description: + - "Switch image from vllm/vllm-openai:v0.20.1 (cu130) to public.ecr.aws/q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 (vLLM nightly release-repo build, cu129)" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/PLACEHOLDER From d12c52e9fa97b3532179d63aa67f44e2071bcb0a Mon Sep 17 00:00:00 2001 From: functionstackx <47992694+functionstackx@users.noreply.github.com> Date: Tue, 5 May 2026 13:44:25 -0400 Subject: [PATCH 2/3] perf-changelog: fill in PR #1284 link Co-Authored-By: Claude Opus 4.7 (1M context) --- perf-changelog.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 6f8a75bcc..3c4c0e688 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -2219,4 +2219,4 @@ - dsv4-fp8-h200-vllm-mtp description: - "Switch image from vllm/vllm-openai:v0.20.1 (cu130) to public.ecr.aws/q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 (vLLM nightly release-repo build, cu129)" - pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/PLACEHOLDER + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1284 From 914853257176c69f8e0077aa2c38f1c338110e3d Mon Sep 17 00:00:00 2001 From: functionstackx <47992694+functionstackx@users.noreply.github.com> Date: Tue, 5 May 2026 13:45:34 -0400 Subject: [PATCH 3/3] dsv4-fp8-h200-vllm-mtp: use enroot-compat URL form (public.ecr.aws#...) Enroot expects the registry and path to be separated by '#' rather than '/' in the image URL. Other entries in nvidia-master.yaml already use this form (nvcr.io#..., ghcr.io#...). Co-Authored-By: Claude Opus 4.7 (1M context) --- .github/configs/nvidia-master.yaml | 2 +- perf-changelog.yaml | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 32d06d54e..e38cab781 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -2634,7 +2634,7 @@ dsv4-fp8-h200-vllm: # (the non-MTP entry above is still on the deepseekv4-cu129 tag) and adds # --speculative-config '{"method":"mtp","num_speculative_tokens":2}'. dsv4-fp8-h200-vllm-mtp: - image: public.ecr.aws/q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 + image: public.ecr.aws#q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 model: deepseek-ai/DeepSeek-V4-Pro model-prefix: dsv4 runner: h200 diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 3c4c0e688..3259cdc26 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -2218,5 +2218,5 @@ - config-keys: - dsv4-fp8-h200-vllm-mtp description: - - "Switch image from vllm/vllm-openai:v0.20.1 (cu130) to public.ecr.aws/q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 (vLLM nightly release-repo build, cu129)" + - "Switch image from vllm/vllm-openai:v0.20.1 (cu130) to public.ecr.aws#q9t5s3a7/vllm-release-repo:fbd51e3dfc902364fddab316ef1337c4f261de1a-x86_64-cu129 (vLLM nightly release-repo build, cu129; enroot URL form with `#` separating registry from path)" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1284