[ROCm][CI] Fine-tuning queues and test names (#39238)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jun 25, 2026 at 15:24 UTC
2a6f8f0c05ab1dd0b11540157fb72b6888883aab
6 files changed
+43
-56
.buildkite/test-amd.yaml
+8
-22
@@ -771,7 +771,7 @@ steps:
771
772
#----------------------------------------------------------- mi300 · cuda ------------------------------------------------------------#
773
774
-- label: Platform Tests (CUDA) # TBD
774
+- label: Platform Tests # TBD
775
timeout_in_minutes: 180
776
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
777
agent_pool: mi300_1
@@ -831,7 +831,7 @@ steps:
831
- pytest -v -s distributed/test_eplb_execute.py
832
- pytest -v -s distributed/test_eplb_spec_decode.py
833
834
-- label: Distributed Tests (2xH100-2xMI250) # TBD
834
+- label: Distributed Tests (2xH100-2xMI300) # TBD
835
timeout_in_minutes: 180
836
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
837
agent_pool: mi300_2
@@ -843,13 +843,19 @@ steps:
843
- vllm/model_executor/layers/fused_moe/
844
- vllm/v1/attention/backends/
845
- vllm/v1/attention/selector.py
846
+ - tests/v1/distributed/test_dbo.py
847
- tests/distributed/test_context_parallel.py
848
- examples/features/data_parallel/data_parallel_offline.py
849
- vllm/_aiter_ops.py
850
- vllm/platforms/rocm.py
851
commands:
852
- pytest -v -s tests/distributed/test_context_parallel.py
853
+ - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py
854
+ - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
855
- VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=allgather_reducescatter --disable-nccl-for-dp-synchronization
856
+ - pytest -v -s tests/v1/distributed/test_dbo.py
857
+ - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
858
+ - pytest -v -s tests/distributed/test_packed_tensor.py
859
860
- label: Distributed Tests (4xA100-4xMI300) # TBD
861
timeout_in_minutes: 180
@@ -2195,26 +2201,6 @@ steps:
2201
- TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
2202
- DP_SIZE=2 pytest -v -s entrypoints/openai/test_multi_api_servers.py
2203
2198
-- label: Distributed Tests (2xH100-2xMI300) # TBD
2199
- timeout_in_minutes: 180
2200
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
2201
- agent_pool: mi300_2
2202
- num_gpus: 2
2203
- working_dir: "/vllm-workspace/"
2204
- source_file_dependencies:
2205
- - vllm/distributed/
2206
- - vllm/v1/distributed/
2207
- - vllm/model_executor/layers/fused_moe/
2208
- - tests/v1/distributed/test_dbo.py
2209
- - vllm/_aiter_ops.py
2210
- - vllm/platforms/rocm.py
2211
- commands:
2212
- - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py
2213
- - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
2214
- - pytest -v -s tests/v1/distributed/test_dbo.py
2215
- - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
2216
- - pytest -v -s tests/distributed/test_packed_tensor.py
2217
-
2204
- label: Metrics, Tracing (2 GPUs) # TBD
2205
timeout_in_minutes: 20
2206
mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
.buildkite/test_areas/cuda.yaml
+2
-2
@@ -2,8 +2,8 @@ group: CUDA
2
depends_on:
3
- image-build
4
steps:
5
-- label: Platform Tests (CUDA)
6
- key: platform-tests-cuda
5
+- label: Platform Tests
6
+ key: platform-tests
7
timeout_in_minutes: 15
8
device: h200_18gb
9
source_file_dependencies:
.buildkite/test_areas/distributed.yaml
+7
-7
@@ -174,8 +174,8 @@ steps:
174
# test multi-node TP with multiproc executor (simulated on single node)
175
- pytest -v -s distributed/test_multiproc_executor.py::test_multiproc_executor_multi_node
176
177
-- label: Distributed Tests (8 GPUs)(H100)
178
- key: distributed-tests-8-gpus-h100
177
+- label: Distributed Tests (8xH100)
178
+ key: distributed-tests-8xh100
179
timeout_in_minutes: 10
180
device: h100
181
num_devices: 8
@@ -195,8 +195,8 @@ steps:
195
# test with torchrun tp=2 and dp=4 with ep
196
- torchrun --nproc-per-node=8 ../examples/features/torchrun/torchrun_dp_example_offline.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep
197
198
-- label: Distributed Tests (4 GPUs)(A100)
199
- key: distributed-tests-4-gpus-a100
198
+- label: Distributed Tests (4xA100)
199
+ key: distributed-tests-4xa100
200
device: a100
201
optional: true
202
num_devices: 4
@@ -211,7 +211,7 @@ steps:
211
- pytest -v -s -x lora/test_mixtral.py
212
213
- label: Distributed Tests (2xH100-2xMI300)
214
- key: distributed-tests-2-gpus-h100
214
+ key: distributed-tests-2xh100-2xmi300
215
timeout_in_minutes: 15
216
device: h100
217
optional: true
@@ -237,8 +237,8 @@ steps:
237
- VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
238
- pytest -v -s tests/distributed/test_packed_tensor.py
239
240
-- label: Distributed Tests (2 GPUs)(B200)
241
- key: distributed-tests-2-gpus-b200
240
+- label: Distributed Tests (2xB200)
241
+ key: distributed-tests-2xb200
242
device: b200-k8s
243
optional: true
244
working_dir: "/vllm-workspace/"
.buildkite/test_areas/e2e_integration.yaml
+6
-6
@@ -2,8 +2,8 @@ group: E2E Integration
2
depends_on:
3
- image-build
4
steps:
5
-- label: DeepSeek V2-Lite Sync EPLB Accuracy
6
- key: deepseek-v2-lite-sync-eplb-accuracy
5
+- label: DeepSeek V2-Lite Sync EPLB Accuracy (4xH100)
6
+ key: deepseek-v2-lite-sync-eplb-accuracy-4xh100
7
timeout_in_minutes: 60
8
device: h100
9
optional: true
@@ -12,8 +12,8 @@ steps:
12
commands:
13
- bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010
14
15
-- label: Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy
16
- key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy
15
+- label: Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy (4xH100)
16
+ key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-4xh100
17
timeout_in_minutes: 60
18
device: h100
19
optional: true
@@ -22,8 +22,8 @@ steps:
22
commands:
23
- bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020
24
25
-- label: Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy (B200)
26
- key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-b200
25
+- label: Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy (2xB200)
26
+ key: qwen3-30b-a3b-fp8-block-sync-eplb-accuracy-2xb200
27
timeout_in_minutes: 60
28
device: b200-k8s
29
optional: true
.buildkite/test_areas/kernels.yaml
+4
-4
@@ -274,8 +274,8 @@ steps:
274
- pytest -v -s kernels/helion/
275
276
277
-- label: Kernels FP8 MoE Test (1 H100)
278
- key: kernels-fp8-moe-test-1-h100
277
+- label: Kernels FP8 MoE Test (1xH100)
278
+ key: kernels-fp8-moe-test-1xh100
279
timeout_in_minutes: 90
280
device: h100
281
num_devices: 1
@@ -291,8 +291,8 @@ steps:
291
- pytest -v -s kernels/moe/test_triton_moe_no_act_mul.py
292
- pytest -v -s kernels/moe/test_triton_moe_ptpc_fp8.py
293
294
-- label: Kernels FP8 MoE Test (2 H100s)
295
- key: kernels-fp8-moe-test-2-h100s
294
+- label: Kernels FP8 MoE Test (2xH100)
295
+ key: kernels-fp8-moe-test-2xh100
296
timeout_in_minutes: 90
297
device: h100
298
num_devices: 2
.buildkite/test_areas/lm_eval.yaml
+16
-15
@@ -28,7 +28,8 @@ steps:
28
- vllm/_aiter_ops.py
29
- vllm/platforms/rocm.py
30
31
-# - label: LM Eval Large Models (4 GPUs)(A100)
31
+# - label: LM Eval Large Models (4xA100)
32
+# key: lm-eval-large-models-4xa100
33
# device: a100
34
# optional: true
35
# num_devices: 4
@@ -40,8 +41,8 @@ steps:
41
# - export VLLM_WORKER_MULTIPROC_METHOD=spawn
42
# - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
43
43
-- label: LM Eval Large Models (4 GPUs)(H100)
44
- key: lm-eval-large-models-4-gpus-h100
44
+- label: LM Eval Large Models (4xH100)
45
+ key: lm-eval-large-models-4xh100
46
device: h100
47
optional: true
48
num_devices: 4
@@ -53,8 +54,8 @@ steps:
54
- export VLLM_USE_DEEP_GEMM=0 # We found Triton is faster than DeepGEMM for H100
55
- pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-hopper.txt --tp-size=4
56
56
-- label: LM Eval Small Models (B200)
57
- key: lm-eval-small-models-b200
57
+- label: LM Eval Small Models (2xB200)
58
+ key: lm-eval-small-models-2xb200
59
timeout_in_minutes: 120
60
device: b200-k8s
61
optional: true
@@ -64,8 +65,8 @@ steps:
65
commands:
66
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt
67
67
-- label: LM Eval Large Models (B200, EP)
68
- key: lm-eval-large-models-b200-ep
68
+- label: LM Eval Large Models EP (2xB200)
69
+ key: lm-eval-large-models-ep-2xb200
70
timeout_in_minutes: 120
71
device: b200-k8s
72
optional: true
@@ -76,8 +77,8 @@ steps:
77
commands:
78
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell-ep.txt
79
79
-- label: LM Eval Qwen3.5 Models (B200)
80
- key: lm-eval-qwen3-5-models-b200
80
+- label: LM Eval Qwen3.5 Models (2xB200)
81
+ key: lm-eval-qwen3-5-models-2xb200
82
timeout_in_minutes: 120
83
device: b200-k8s
84
optional: true
@@ -93,8 +94,8 @@ steps:
94
commands:
95
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-blackwell.txt
96
96
-- label: LM Eval Large Models (H200)
97
- key: lm-eval-large-models-h200
97
+- label: LM Eval Large Models (8xH200)
98
+ key: lm-eval-large-models-8xh200
99
timeout_in_minutes: 60
100
device: h200
101
optional: true
@@ -192,8 +193,8 @@ steps:
193
commands:
194
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant.txt
195
195
-- label: GPQA Eval (GPT-OSS) (H100)
196
- key: gpqa-eval-gpt-oss-h100
196
+- label: GPQA Eval (GPT-OSS) (2xH100)
197
+ key: gpqa-eval-gpt-oss-2xh100
198
timeout_in_minutes: 120
199
device: h100
200
optional: true
@@ -206,8 +207,8 @@ steps:
207
- uv pip install --system 'gpt-oss[eval]==0.0.5'
208
- pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-h100.txt
209
209
-- label: GPQA Eval (GPT-OSS) (B200)
210
- key: gpqa-eval-gpt-oss-b200
210
+- label: GPQA Eval (GPT-OSS) (2xB200)
211
+ key: gpqa-eval-gpt-oss-2xb200
212
timeout_in_minutes: 120
213
device: b200-k8s
214
optional: true