-
Notifications
You must be signed in to change notification settings - Fork 61
Add Slurm multi-node and WideEP support for AI Dynamo #1018
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Open
juntaowww
wants to merge
7
commits into
NVIDIA:main
Choose a base branch
from
juntaowww:feature/ai-dynamo-wideep
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
Changes from all commits
Commits
Show all changes
7 commits
Select commit
Hold shift + click to select a range
5c8f79d
add multinode AI Dynamo workers
juntaowww f9bf940
fix AI Dynamo disaggregation mode handling
juntaowww accae4a
fix SGLang multinode worker hangs
juntaowww 8b1547f
fix AI Dynamo copyright headers
juntaowww 493c781
add AI Dynamo WideEP support
juntaowww ac2e024
harden AI Dynamo WideEP orchestration
juntaowww e5cfcd5
harden AI Dynamo WideEP runtime handling
juntaowww File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
48 changes: 48 additions & 0 deletions
48
conf/experimental/ai_dynamo/test_scenario/sglang_multinode_worker_slurm.toml
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,48 @@ | ||
| # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES | ||
| # Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| name = "dynamo-sglang-multinode-worker-slurm" | ||
| job_status_check = false | ||
|
|
||
| [[Tests]] | ||
| id = "test.disagg.multinode-worker" | ||
| test_name = "sglang" | ||
| time_limit = "00:20:00" | ||
|
|
||
| [Tests.cmd_args] | ||
| [Tests.cmd_args.dynamo] | ||
| model = "Qwen/Qwen3-0.6B" | ||
| node-setup-cmd = "hostname" | ||
|
|
||
| # Work around the known SGLang multi-node TP overlap-scheduler hang: | ||
| # https://github.com/sgl-project/sglang/issues/26454 | ||
| [Tests.cmd_args.dynamo.prefill_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| extra-args = "--trust-remote-code --skip-tokenizer-init --enable-metrics --disable-overlap-schedule" | ||
|
|
||
| [Tests.cmd_args.dynamo.prefill_worker.args] | ||
| tensor-parallel-size = 8 | ||
|
|
||
| # Work around the known SGLang multi-node TP CUDA graph replay hang: | ||
| # https://github.com/sgl-project/sglang/issues/29548 | ||
| [Tests.cmd_args.dynamo.decode_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| extra-args = "--trust-remote-code --skip-tokenizer-init --enable-metrics --disable-overlap-schedule --disable-cuda-graph" | ||
|
|
||
| [Tests.cmd_args.dynamo.decode_worker.args] | ||
| tensor-parallel-size = 8 |
68 changes: 68 additions & 0 deletions
68
conf/experimental/ai_dynamo/test_scenario/sglang_wideep_slurm.toml
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,68 @@ | ||
| # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES | ||
| # Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| name = "dynamo-sglang-qwen3-30b-a3b-fp8-wideep-slurm" | ||
| job_status_check = false | ||
|
|
||
| [[Tests]] | ||
| id = "test.disagg.1p1d.multinode-wideep" | ||
| test_name = "sglang" | ||
| time_limit = "00:30:00" | ||
|
|
||
| [Tests.cmd_args] | ||
| [Tests.cmd_args.dynamo] | ||
| model = "Qwen/Qwen3-30B-A3B-FP8" | ||
|
juntaowww marked this conversation as resolved.
|
||
| node-setup-cmd = "hostname" | ||
|
|
||
| # Work around the SGLang multi-node overlap-scheduler hang: | ||
| # https://github.com/sgl-project/sglang/issues/26454 | ||
| [Tests.cmd_args.dynamo.prefill_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| extra-args = "--trust-remote-code --skip-tokenizer-init --enable-metrics --enable-dp-attention --disable-overlap-schedule" | ||
|
|
||
| [Tests.cmd_args.dynamo.prefill_worker.args] | ||
| model-path = "Qwen/Qwen3-30B-A3B-FP8" | ||
| served-model-name = "Qwen/Qwen3-30B-A3B-FP8" | ||
| tensor-parallel-size = 8 | ||
| pipeline-parallel-size = 1 | ||
| data-parallel-size = 8 | ||
| ep-size = 8 | ||
| moe-a2a-backend = "deepep" | ||
| deepep-mode = "normal" | ||
| mem-fraction-static = 0.75 | ||
| load-balance-method = "round_robin" | ||
| watchdog-timeout = 3600 | ||
|
|
||
| # Work around the SGLang multi-node overlap-scheduler and CUDA graph replay hangs: | ||
| # https://github.com/sgl-project/sglang/issues/26454 | ||
| # https://github.com/sgl-project/sglang/issues/29548 | ||
| [Tests.cmd_args.dynamo.decode_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| extra-args = "--trust-remote-code --skip-tokenizer-init --enable-metrics --enable-dp-attention --prefill-round-robin-balance --disable-overlap-schedule --disable-cuda-graph" | ||
|
|
||
| [Tests.cmd_args.dynamo.decode_worker.args] | ||
| model-path = "Qwen/Qwen3-30B-A3B-FP8" | ||
| served-model-name = "Qwen/Qwen3-30B-A3B-FP8" | ||
| tensor-parallel-size = 8 | ||
| pipeline-parallel-size = 1 | ||
| data-parallel-size = 8 | ||
| ep-size = 8 | ||
| moe-a2a-backend = "deepep" | ||
| deepep-mode = "normal" | ||
| mem-fraction-static = 0.75 | ||
| watchdog-timeout = 3600 | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
54 changes: 54 additions & 0 deletions
54
conf/experimental/ai_dynamo/test_scenario/vllm_multinode_worker_slurm.toml
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,54 @@ | ||
| # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES | ||
| # Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| name = "dynamo-vllm-multinode-worker-slurm" | ||
| job_status_check = false | ||
|
|
||
| [[Tests]] | ||
| id = "test.disagg.multinode-worker" | ||
| test_name = "vLLM" | ||
| time_limit = "00:10:00" | ||
|
|
||
| [Tests.cmd_args] | ||
| [Tests.cmd_args.dynamo.dcgm_exporter] | ||
| enabled = true | ||
| docker-image-url = "nvcr.io/nvidia/k8s/dcgm-exporter:4.5.2-4.8.1-distroless" | ||
|
|
||
| [Tests.cmd_args.dynamo.prefill_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| [Tests.cmd_args.dynamo.prefill_worker.args] | ||
| pipeline-parallel-size = 1 | ||
|
|
||
| [Tests.cmd_args.dynamo.decode_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| [Tests.cmd_args.dynamo.decode_worker.args] | ||
| pipeline-parallel-size = 1 | ||
|
|
||
| [[Tests.cmd_args.aiperf_phases]] | ||
| name = "round_1" | ||
| [Tests.cmd_args.aiperf_phases.args] | ||
| concurrency = 4 | ||
| request-count = 50 | ||
| server-metrics = "auto" | ||
|
|
||
| [[Tests.cmd_args.aiperf_phases]] | ||
| name = "round_2" | ||
| [Tests.cmd_args.aiperf_phases.args] | ||
| concurrency = 8 | ||
| request-count = 50 | ||
| server-metrics = "auto" |
55 changes: 55 additions & 0 deletions
55
conf/experimental/ai_dynamo/test_scenario/vllm_wideep_slurm.toml
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,55 @@ | ||
| # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES | ||
| # Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| name = "dynamo-vllm-qwen3-30b-a3b-wideep-slurm" | ||
| job_status_check = false | ||
|
|
||
| [[Tests]] | ||
| id = "test.disagg.1p1d.multinode-wideep" | ||
| test_name = "vLLM" | ||
| time_limit = "00:30:00" | ||
|
|
||
| [Tests.cmd_args] | ||
| [Tests.cmd_args.dynamo] | ||
| model = "Qwen/Qwen3-30B-A3B" | ||
| node-setup-cmd = "hostname" | ||
|
|
||
| # On four-GPU nodes, TP=1 and DP=8 place four DP ranks per node. | ||
| # vLLM derives EP=TP*DP=8, so each role's expert group spans both nodes. | ||
| [Tests.cmd_args.dynamo.prefill_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| extra-args = "--enable-expert-parallel" | ||
|
|
||
| [Tests.cmd_args.dynamo.prefill_worker.args] | ||
| tensor-parallel-size = 1 | ||
| pipeline-parallel-size = 1 | ||
| data-parallel-size = 8 | ||
| all2all-backend = "allgather_reducescatter" | ||
| max-model-len = 8192 | ||
| kv-events-config = '{"publisher":"zmq","topic":"kv-events","endpoint":"tcp://*:20080","enable_kv_cache_events":true}' | ||
|
|
||
| [Tests.cmd_args.dynamo.decode_worker] | ||
| num-nodes = 2 | ||
| nodes-per-worker = 2 | ||
| extra-args = "--enable-expert-parallel" | ||
|
|
||
| [Tests.cmd_args.dynamo.decode_worker.args] | ||
| tensor-parallel-size = 1 | ||
| pipeline-parallel-size = 1 | ||
| data-parallel-size = 8 | ||
| all2all-backend = "allgather_reducescatter" | ||
| max-model-len = 8192 |
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
please keep the existing configs as is. only add new scenarios pointing to the existing test tomls, overriding specific values