From 6be7261d37a65ee5b3a5a28bbb08313da24f17ae Mon Sep 17 00:00:00 2001 From: Chenhan Yu Date: Mon, 15 Jun 2026 12:32:23 -0700 Subject: [PATCH] =?UTF-8?q?launcher:=20Nemotron-120B=20specdec=5Fbench=20?= =?UTF-8?q?=E2=80=94=20kill=20stale=20=5Fcells/=20reference=20(post-PR-#15?= =?UTF-8?q?64)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The doc-comment header on this parent YAML still showed the pre-PR-#1564 invocation pattern: pipeline.task_0.args+=["--runtime_params common/specdec_bench/_cells/.yaml", ...] PR #1564 removed the _cells/ and _runtime_params/ directories. Cell- specific knobs are now CLI overrides at slurm-invoke time, not files. But this doc-comment kept advertising the old pattern, and multiple pensieve-intern agent dispatches across OMNIML-5024 / 5025 / 5026 / 5027 keep authoring new _cells/.yaml files on the (gemma-4, MTP, vLLM) Epic — exactly because they're scanning launcher/examples/ for a reference invocation and finding this one. The cell SPEC says "no per-cell YAML file" five times in prose, but prose loses to a concrete checked-in counter-example. Rewriting the doc-comment to the post-#1564 pattern matches the canonical Qwen3.5-4B reference template that cell.md tells the agent to model from. No code/yaml change — comment-only. Signed-off-by: Chenhan Yu --- .../specdec_bench_dflash_vllm.yaml | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/tools/launcher/examples/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/specdec_bench_dflash_vllm.yaml b/tools/launcher/examples/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/specdec_bench_dflash_vllm.yaml index 5bf06cb709d..18b0db85c5a 100644 --- a/tools/launcher/examples/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/specdec_bench_dflash_vllm.yaml +++ b/tools/launcher/examples/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/specdec_bench_dflash_vllm.yaml @@ -10,10 +10,15 @@ # H100/A100. Surfaced on OMNIML-4969: tp_size=2 OOMed at model load # (~80 GB needed per GPU at tp=2, only 79.11 GiB available). # -# Slurm run on cw_dfw — cells override per-cell runtime_params, -# --save_dir, --block_size, --num_requests via pipeline.task_N.args+=[...]: +# Slurm run on cw_dfw — cells override per-cell knobs via +# pipeline.task_N.args+=[...] CLI flags (post-PR-#1564 convention; no +# per-cell runtime_params file, no _cells/.yaml): # -# uv run slurm.py # --yaml modules/Model-Optimizer/tools/launcher/examples/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/specdec_bench_dflash_vllm.yaml # --yes detach=true # pipeline.task_0.args+=["--runtime_params common/specdec_bench/_cells/.yaml","--save_dir /scratchspace//qualitative","--block_size 4"] # pipeline.task_1.args+=["--runtime_params common/specdec_bench/_cells/.yaml","--save_dir /scratchspace//throughput_32k","--num_requests 80","--block_size 4"] +# uv run slurm.py \ +# --yaml modules/Model-Optimizer/tools/launcher/examples/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/specdec_bench_dflash_vllm.yaml \ +# --yes detach=true \ +# pipeline.task_0.args+=["--save_dir /scratchspace//qualitative","--block_size 4"] \ +# pipeline.task_1.args+=["--save_dir /scratchspace//throughput_32k","--num_requests 80","--block_size 4"] # # Reference run: cicd_1781024226 (cw_dfw) produced # qualitative Average_AL = 2.7316 (11-cat breakdown)