test: add qwen3 cases

[ruodil]

PR title

Please write the PR title by following template:

[JIRA ticket link/nvbug link/github issue link][fix/feat/doc/infra/...] <summary of this PR>

For example, assume I have a PR hope to support a new feature about cache manager of Jira TRTLLM-1000 ticket, it would be like

[TRTLLM-1000][feat] Support a new feature about cache manager

Description

1. add qwen3 test cases for 8xB200, 4xB200, 8xH20 and 8xH200
2. set default pytorch config for qwen3
3. waive some failed cases and fix some case name issue

jenkins job of qwen3 running on 8xB200 and 4xB200: https://prod.blsm.nvidia.com/swqa-tensorrt-qa-test/view/TRT-LLM-Perf-Pipelines/job/TRT-LLM-Perf-Test-Cluster/270/
case name:
qwen3_235b_a22b_fp4-bench-pytorch-float4-input_output_len:1000,2000-con:512-ep:4-gpus:4
pytorch config:{'print_iter_log': True, 'use_cuda_graph': True, 'cuda_graph_padding_enabled': True, 'enable_attention_dp': True}
perf value:

= WORLD + RUNTIME INFORMATION

18:38:46 TP Size: 4
18:38:46 PP Size: 1
18:38:46 EP Size: 4
18:38:46 Max Runtime Batch Size: 512
18:38:46 Max Runtime Tokens: 2048
18:38:46 Scheduling Policy: GUARANTEED_NO_EVICT
18:38:46 KV Memory Percentage: 90.00%
18:38:46 Issue Rate (req/sec): 6.4743E+13
18:38:46

= PERFORMANCE OVERVIEW

18:38:46 Request Throughput (req/sec): 5.0949
18:38:46 Total Output Throughput (tokens/sec): 10189.7835
18:38:46 Total Token Throughput (tokens/sec): 15284.6752
18:38:46 Total Latency (ms): 100492.8125
18:38:46 Average request latency (ms): 98531.9329
18:38:46 Per User Output Throughput [w/ ctx] (tps/user): 20.3006
18:38:46 Per GPU Output Throughput (tps/gpu): 2547.4459
18:38:46
18:38:46 -- Request Latency Breakdown (ms) -----------------------
18:38:46
18:38:46 [Latency] P50 : 98457.1125
18:38:46 [Latency] P90 : 100137.7227
18:38:46 [Latency] P95 : 100305.4983
18:38:46 [Latency] P99 : 100435.9094
18:38:46 [Latency] MINIMUM: 96809.5739
18:38:46 [Latency] MAXIMUM: 100459.7999
18:38:46 [Latency] AVERAGE: 98531.9329

case name: qwen3_235b_a22b_fp4-bench-pytorch-float4-maxbs:512-maxnt:2048-input_output_len:1000,2000-con:8-ep:8-gpus:8
pytorch config: {'print_iter_log': True, 'use_cuda_graph': True, 'cuda_graph_padding_enabled': True, 'enable_attention_dp': False, 'moe_backend': 'TRTLLM'}
perf value:

= PYTORCH BACKEND

19:09:04 Model: qwen3_235b_a22b_fp4_hf
19:09:04 Model Path: /llm-models/Qwen3/saved_models_Qwen3-235B-A22B_nvfp4_hf
19:09:04 TensorRT-LLM Version: 0.21.0rc2
19:09:04 Dtype: bfloat16
19:09:04 KV Cache Dtype: FP8
19:09:04 Quantization: NVFP4
19:09:04

= REQUEST DETAILS

19:09:04 Number of requests: 512
19:09:04 Number of concurrent requests: 7.9998
19:09:04 Average Input Length (tokens): 1000.0000
19:09:04 Average Output Length (tokens): 2000.0000

= WORLD + RUNTIME INFORMATION

19:09:04 TP Size: 8
19:09:04 PP Size: 1
19:09:04 EP Size: 8
19:09:04 Max Runtime Batch Size: 512
19:09:04 Max Runtime Tokens: 2048
19:09:04 Scheduling Policy: GUARANTEED_NO_EVICT
19:09:04 KV Memory Percentage: 90.00%
19:09:04 Issue Rate (req/sec): 3.2525E+18
19:09:04

= PERFORMANCE OVERVIEW

19:09:04 Request Throughput (req/sec): 0.3028
19:09:04 Total Output Throughput (tokens/sec): 605.5911
19:09:04 Total Token Throughput (tokens/sec): 908.3867
19:09:04 Total Latency (ms): 1690909.7974
19:09:04 Average request latency (ms): 26419.9106
19:09:04 Per User Output Throughput [w/ ctx] (tps/user): 75.7018
19:09:04 Per GPU Output Throughput (tps/gpu): 75.6989
19:09:04
19:09:04 -- Request Latency Breakdown (ms) -----------------------
19:09:04
19:09:04 [Latency] P50 : 26422.4362
19:09:04 [Latency] P90 : 26521.9194
19:09:04 [Latency] P95 : 26568.6153
19:09:04 [Latency] P99 : 26613.1254
19:09:04 [Latency] MINIMUM: 25654.9526
19:09:04 [Latency] MAXIMUM: 27021.6187
19:09:04 [Latency] AVERAGE: 26419.9106

jenkins job of qwen3 running on 8xH20: https://prod.blsm.nvidia.com/swqa-tensorrt-qa-test/view/TRT-LLM-Perf-Pipelines/job/Debug-TRT-LLM-Perf-Test/94/
case name: qwen3_235b_a22b_fp8-bench-pytorch-float8-input_output_len:1000,2000-con:256-ep:8-gpus:8
pytorch config: {'print_iter_log': True, 'use_cuda_graph': True, 'cuda_graph_padding_enabled': True, 'enable_attention_dp': True}
perf value:

= PYTORCH BACKEND
Model: qwen3_235b_a22b_fp8_hf
Model Path: /code/llm_models/Qwen3/saved_models_Qwen3-235B-A22B_fp8_hf
TensorRT-LLM Version: 0.21.0rc2
Dtype: bfloat16
KV Cache Dtype: FP8
Quantization: FP8

= REQUEST DETAILS

Number of requests: 512
Number of concurrent requests: 255.6332
Average Input Length (tokens): 1000.0000
Average Output Length (tokens): 2000.0000

= WORLD + RUNTIME INFORMATION

TP Size: 8
PP Size: 1
EP Size: 8
Max Runtime Batch Size: 512
Max Runtime Tokens: 2048
Scheduling Policy: GUARANTEED_NO_EVICT
KV Memory Percentage: 90.00%
Issue Rate (req/sec): 1.6205E+17

= PERFORMANCE OVERVIEW

Request Throughput (req/sec): 3.5417
Total Output Throughput (tokens/sec): 7083.4665
Total Token Throughput (tokens/sec): 10625.1998
Total Latency (ms): 144561.9873
Average request latency (ms): 72177.4335
Per User Output Throughput [w/ ctx] (tps/user): 27.9136
Per GPU Output Throughput (tps/gpu): 885.4333

-- Request Latency Breakdown (ms) -----------------------

[Latency] P50 : 72617.8651
[Latency] P90 : 80713.5184
[Latency] P95 : 82193.1687
[Latency] P99 : 82933.9907
[Latency] MINIMUM: 61589.9742
[Latency] MAXIMUM: 82934.7629
[Latency] AVERAGE: 72177.4335

Test Coverage

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

run [--disable-fail-fast --skip-test --stage-list "A10-1, xxx" --gpu-type "A30, H100_PCIe" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-[Post-Merge]-1, xxx"]

Launch build/test pipelines. All previously running jobs will be killed.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests. Will also run L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-[Post-Merge]-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-[Post-Merge]-1, xxx".

For guidance on mapping tests to stage names, see docs/source/reference/ci-overview.md.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

[byshiue]

From the log of case name: qwen3_235b_a22b_fp4-bench-pytorch-float4-maxbs:512-maxnt:2048-input_output_len:1000,2000-con:8-ep:8-gpus:8, I don't see that we enable the trtllm mode backend. Can we double check?

[ruodil]

From the log of case name: qwen3_235b_a22b_fp4-bench-pytorch-float4-maxbs:512-maxnt:2048-input_output_len:1000,2000-con:8-ep:8-gpus:8, I don't see that we enable the trtllm mode backend. Can we double check?

yes, I added trtllm moe backend in this case and start another test, will update the log once it finished.

[byshiue]

/bot skip --comment "only update the perf list of qa"

[tensorrt-cicd]

PR_Github #9452 [ skip ] triggered by Bot

[tensorrt-cicd]

PR_Github #9452 [ skip ] completed with state SUCCESS
Skipping testing for commit 3c41e90