NVIDIA/srt-slurm v1.0.0

v1.0.0

Repository: NVIDIA/srt-slurm

Tag: v1.0.0

Published: 2026-06-05T23:10:11Z

Prerelease: no

Release notes:

What's Changed

• Update README to include TensorRT LLM and vLLM in description by @nlevin-ui in https://github.com/NVIDIA/srt-slurm/pull/1
• [MISC] Add License / headers, and a small check to prepare for release by @xinli-sw in https://github.com/NVIDIA/srt-slurm/pull/4
• feat: enable runtime container detection for portable dynamo source builds by @qiching in https://github.com/NVIDIA/srt-slurm/pull/3
• Sync ishandhanani/srt-slurm history into NVIDIA/srt-slurm by @csahithi in https://github.com/NVIDIA/srt-slurm/pull/14
• Add trace-replay benchmark type by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/16
• fix: use custom_tokenizer to workaround the trtllm + glm5 tokenizer loading issue by @richardhuo-nv in https://github.com/NVIDIA/srt-slurm/pull/20
• fix: add nvidia pypi as an extra index to be able to pip install the prerelease dynamo wheels by @richardhuo-nv in https://github.com/NVIDIA/srt-slurm/pull/22
• fix: support cross-arch clusters (x86_64 login, aarch64 compute) by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/17
• feat: trace-replay benchmark with aiperf_args passthrough by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/18
• feat: add mocker backend for pipeline smoke tests by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/25
• feat: separate login-node and compute-node venvs by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/29
• feat: runtime fingerprinting, identity verification, and lockfile by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/19
• feat: configurable NATS max_payload for disagg serving by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/31
• Copy {job_id}.json into log directory for S3 upload by @KaunilD in https://github.com/NVIDIA/srt-slurm/pull/15
• TRTLLM nsys profiling harness + Dynamo OTEL tracing automation by @karen-sy in https://github.com/NVIDIA/srt-slurm/pull/27
• Add CODEOWNERS file by @xinli-sw in https://github.com/NVIDIA/srt-slurm/pull/37
• Add CSV export for sa-bench rollup by @weireweire in https://github.com/NVIDIA/srt-slurm/pull/26
• Sanitize srun output in node IP resolution by @weireweire in https://github.com/NVIDIA/srt-slurm/pull/38
• feat: lockfile v2 — shareable recipe + lock section by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/32
• fix: Install maturin if not present by @trevor-m in https://github.com/NVIDIA/srt-slurm/pull/45
• [codex] Add generic telemetry and custom benchmark support by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/43
• [codex] Port HF cache cleanup by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/49
• Add srt-slurm MCP spec server and preflight validation by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/53
• Push logs_url to status API eagerly and via final PUT by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/54
• [codex] narrow srtctl mcp to authoring and validation by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/55
• [codex] Keep MCP validation off host cluster config by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/56
• fix: emit aggregated resources and harden sa-bench rollup by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/58
• feat: use pre-generated custom dataset for benchmarking MTP with chat template by @richardhuo-nv in https://github.com/NVIDIA/srt-slurm/pull/64
• docs: loud warnings on custom benchmark templating and nginx-off mode by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/66
• feat(sa-bench): add sglang DeepSeek-V4 tokenizer by @YAMY1234 in https://github.com/NVIDIA/srt-slurm/pull/73
• feat: DeepSeek-V4-Pro perf recipes for GB300 / GB200 (1k/1k agg) by @elvischenv in https://github.com/NVIDIA/srt-slurm/pull/70
• fix(orchestrate): robust container bootstrap (maturin/protoc/venv-race) by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/81
• fix(sa-bench): actionable error + warmup parity for use_chat_template by @YAMY1234 in https://github.com/NVIDIA/srt-slurm/pull/76
• feat(schema): make gsm8k a first-class BenchmarkType by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/82
• [codex] add AIME benchmark by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/83
• feat(aime): rework around ns eval for reasoning-model parity by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/87
• Add scripts for wideEP; Note we can reach a PD balance with dep8, cc=2048 by @samuellees in https://github.com/NVIDIA/srt-slurm/pull/52
• Revert "Add scripts for wideEP; Note we can reach a PD balance with dep8, cc=2048" by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/89
• refactor(aime): drop structured runner, ship configs/aime/{run.sh,rescore.py} by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/91
• Add the chat template to the glm5 tokenizer and apply that when sampling the requests by @richardhuo-nv in https://github.com/NVIDIA/srt-slurm/pull/65
• feat(config): resolve container aliases for telemetry + preflight by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/101
• [codex] Add Dynamo nightly wheel install support by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/99
• feat(dynamo): cache hash-pinned source builds on /configs by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/88
• Add DeepSeek V4 Pro vLLM GB200 recipes by @alec-flowers in https://github.com/NVIDIA/srt-slurm/pull/102
• feat(config): cluster-wide default_bash_preamble for ulimits and the like by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/104
• fix(nginx): raise file descriptor limit for nginx workers by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/108
• log: always set dyn skip log fmt by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/109
• [NOT FINAL] add wip DSv4 aggregate and disaggregate recipes by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/85
• nginx: rework to make ulimit optional by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/110
• log: demote per-srun command line to DEBUG by @cquil11 in https://github.com/NVIDIA/srt-slurm/pull/111
• fix: using a setup script to install pip in trtllm venv by @richardhuo-nv in https://github.com/NVIDIA/srt-slurm/pull/116
• default dyn log by @ishandhanani in https://github.com/NVIDIA/srt-slurm/pull/118
• feat: Add live monitor to…