test(perf): Extend the Llama-Nemotron-Nano-8B perf-integration-tests (pyt) #4407
Conversation
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
venkywonka
requested review from
Copilot,
ruodil,
tijyojwad,
schetlur-nv and
LarryXFly
There was a problem hiding this comment.
Pull Request Overview
This PR expands end-to-end performance test coverage for the llama_v3.1_nemotron_nano_8b engine on the PyTorch backend, evaluating low and high concurrency patterns across various input/output lengths.
- Adds 8 new pytest entries under a new “torch backend” section for both low (concurrency=1, requests=8) and high (concurrency=250, requests=500) loads.
- Removes two outdated PyTorch backend tests using default input lengths.
- Ensures max batch size is set to 512 in all new scenarios.
Comments suppressed due to low confidence (1)
tests/integration/test_lists/qa/trt_llm_release_perf_test.yml:26
- [nitpick] Consider renaming the section comment '# torch backend' to '# pytorch backend' for consistency and clarity in labeling.
# torch backend
tijyojwad
approved these changes
May 19, 2025
ruodil
approved these changes
May 20, 2025
LarryXFly
approved these changes
May 21, 2025
venkywonka
force-pushed
the
user/venky/ll-nemo-nano-pyt-perf-ext
branch
from
a94c959 to
e8b3fee
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #6156 [ run ] triggered by Bot |
|
PR_Github #6156 [ run ] completed with state |
Signed-off-by: Venky <[email protected]>
This is because the test harness default to no prefill chunking, that means the isl specified is the true context. When explicitly unspecified in the test harness, the `maxnt` passed down to `trtllm-bench` is 2048. This means trtllm-bench gets conflicting inputs when isl>2048 but maxnt=2048; hence overriding maxnt to be consistent with isl for such cases. Signed-off-by: Venky <[email protected]>
venkywonka
force-pushed
the
user/venky/ll-nemo-nano-pyt-perf-ext
branch
from
e8b3fee to
67260d3
Compare
LarryXFly
approved these changes
May 23, 2025
amirkl94
pushed a commit
to amirkl94/TensorRT-LLM
that referenced
this pull request
May 28, 2025
…(pyt) (NVIDIA#4407) * extend pyt nano tests perf coverage Signed-off-by: Venky <[email protected]> * explicitly set maxnt for some cases This is because the test harness default to no prefill chunking, that means the isl specified is the true context. When explicitly unspecified in the test harness, the `maxnt` passed down to `trtllm-bench` is 2048. This means trtllm-bench gets conflicting inputs when isl>2048 but maxnt=2048; hence overriding maxnt to be consistent with isl for such cases. Signed-off-by: Venky <[email protected]> --------- Signed-off-by: Venky <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Expand PyT
llama_v3.1_nemotron_nano_8bperf tests coverageDescription
This PR adds end-to-end performance results for the llama_v3.1_nemotron_nano_8b bfloat16 engine on 1 H100.
Two broad load patterns were evaluated on PyT backend for various ISL/OSL combos:
concurrency = 1,requests = 8concurrency = 250,requests = 500All tests use
max_batch_size = 512.Performance Summary
(req/s)
(tps/gpu)
(ms)
NOTE: the above numbers were generated with prefill chunking disabled (which is the default behavior)