|
293 | 293 |
|
294 | 294 | # TODO(CJQ): make first class upon srt-slurm upstream refactor |
295 | 295 | if [[ "$IS_AGENTIC" == "1" ]]; then |
296 | | - # DSV4 AgentX DEP recipes require NVIDIA/srt-slurm#90's per-node DP launch |
297 | | - # mode so each process owns the four local DP ranks in one CUDA namespace. |
298 | | - # The pinned commit also provides everything else these recipes need: |
299 | | - # - BenchmarkType.CUSTOM + benchmark.command + benchmark.env |
300 | | - # (the hook that hands off to benchmarks/multi_node/agentic_srt.sh) |
301 | | - # - DynamoConfig.wheel (recipes pin the ai-dynamo wheel) |
302 | | - # - mooncake_kv_store for the external AgentX prefix-cache service |
303 | | - # - srtctl apply --no-preflight (model path /mnt/numa1 is compute-node |
304 | | - # local NVMe, invisible to the login-node runner) |
305 | | - # - benchmark_stage srun_options propagation (container-remap-root |
306 | | - # must reach the agentic_srt.sh srun) |
307 | | - git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR" |
308 | | - cd "$SRT_REPO_DIR" |
309 | | - git checkout 975e1e60d1442f42384b439b80dbaa686498b4f3 |
310 | | - mkdir -p recipes/vllm/deepseek-v4/agentic |
| 296 | + # v1.0.27 is the last release with the compatible mooncake_master command; |
| 297 | + # v1.0.28 introduced the unsupported --nof_* flag. |
| 298 | + # The pinned release also provides the vLLM mooncake_kv_store |
| 299 | + SRT_SLURM_AGENTIC_SHA="f6eb42aee4664207dcf2ec601e3bd57bd527efd6" |
| 300 | + git clone --branch v1.0.27 --depth 1 https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR" || exit 1 |
| 301 | + cd "$SRT_REPO_DIR" || exit 1 |
| 302 | + if [[ "$(git rev-parse HEAD)" != "$SRT_SLURM_AGENTIC_SHA" ]]; then |
| 303 | + echo "Error: NVIDIA/srt-slurm v1.0.27 did not resolve to $SRT_SLURM_AGENTIC_SHA" >&2 |
| 304 | + exit 1 |
| 305 | + fi |
| 306 | + mkdir -p recipes/vllm/deepseek-v4/agentic || exit 1 |
311 | 307 | cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic" \ |
312 | | - recipes/vllm/deepseek-v4/agentic |
| 308 | + recipes/vllm/deepseek-v4/agentic || exit 1 |
313 | 309 | elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "dsv4" ]]; then |
314 | 310 | git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR" |
315 | 311 | cd "$SRT_REPO_DIR" |
|
0 commit comments