Skip to content

Change sampling tests with low vocab_size to set top_k equal to vocab_size because default top_k exceeds vocab_size #2486

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.

Sign up for GitHub

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

Closed
Hrishith Thadicherla (hthadicherla) wants to merge 2 commits into from
Closed

Change sampling tests with low vocab_size to set top_k equal to vocab_size because default top_k exceeds vocab_size #2486

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e559fce
Allow top_k greater than vocab_size
hthadicherla Aug 28, 2026
d721cf3
Set top_k to vocab_size in NvTensorRtRtx top-p tests
hthadicherla Aug 31, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions test/sampling_tests.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -823,6 +823,9 @@ TEST(SamplingTests, BatchedSamplingTopPNvTensorRtRtx) {
0.1f, 0.1f, 0.1f, 0.1f, 0.6f};

setup.params->SetSearchOption("top_p", 0.25f);
// top_k defaults to 50, which exceeds this test's overlaid vocab_size and is rejected as an
// invalid request. Set top_k = vocab_size to express "consider all tokens" for pure top-p.
setup.params->SetSearchOption("top_k", vocab_size);

auto generator = OgaGenerator::Create(*setup.model, *setup.params);
generator->SetLogits(*OgaTensor::Create(logits_cpu.data(), std::array<int64_t, 2>{batch_size, vocab_size}));
Expand Down Expand Up @@ -901,6 +904,9 @@ TEST(SamplingTests, RandomizedSamplingTopPNvTensorRtRtx) {
}

setup.params->SetSearchOption("top_p", p);
// top_k defaults to 50, which exceeds this test's overlaid vocab_size and is rejected as an
// invalid request. Set top_k = vocab_size to express "consider all tokens" for pure top-p.
setup.params->SetSearchOption("top_k", vocab_size);

std::random_device rd;
std::mt19937 engine(rd());
Expand Down