[TRTLLM-13781][chore] Remove legacy TensorRT examples#15763
Merged
Wanli-Jiang merged 3 commits intoJul 2, 2026
Merged
Conversation
Wanli-Jiang
force-pushed
the
user/williamj/deprecated-trt-backend-step1p1
branch
from
June 30, 2026 07:45
c1e1c13 to
61717fd
Compare
Wanli-Jiang
marked this pull request as ready for review
June 30, 2026 07:47
Wanli-Jiang
requested review from
QiJune,
Shixiaowei02,
chang-l,
nv-guomingz and
yechank-nvidia
June 30, 2026 07:47
nv-guomingz
approved these changes
Jun 30, 2026
Collaborator
Author
|
/bot run --disable-fail-fast |
Contributor
📝 WalkthroughWalkthroughRemoves all legacy example content across speculative decoding examples ( Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Collaborator
|
PR_Github #56554 [ run ] triggered by Bot. Commit: |
juney-nvidia
approved these changes
Jun 30, 2026
juney-nvidia
left a comment
juney-nvidia
left a comment
Collaborator
There was a problem hiding this comment.
Approved from OSS compliance perspective.
1 task
Wanli-Jiang
enabled auto-merge (squash)
June 30, 2026 08:19
Collaborator
|
PR_Github #56554 [ run ] completed with state
|
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #56611 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #56611 [ run ] completed with state
|
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #56789 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #56789 [ run ] completed with state
|
Wanli-Jiang
force-pushed
the
user/williamj/deprecated-trt-backend-step1p1
branch
2 times, most recently
from
July 1, 2026 04:19
c637b5c to
24223a9
Compare
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #57058 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #57057 [ run ] completed with state |
Collaborator
|
PR_Github #57058 [ run ] completed with state
|
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #57121 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #57121 [ run ] completed with state
|
Collaborator
Author
|
/bot skip --comment "only one case failed and it was passed previously and locally, we can skip checking again and again" |
Collaborator
|
PR_Github #57149 [ skip ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #57149 [ skip ] completed with state |
evezhier
pushed a commit
to evezhier/TensorRT-LLM
that referenced
this pull request
Jul 2, 2026
Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
1 task
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 3, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove two orphaned platform waives (GH200/sm100 video-neva) whose only test entry is now gone. Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 3, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove two orphaned platform waives (GH200/sm100 video-neva) whose only test entry is now gone. Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 3, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove orphaned waives whose only test entry is now gone: the GH200/sm100 video-neva platform waives, and the sm100 "Disable for Blackwell" combined-job waive (its multi-target l0_a100 tensorrt job was removed with the TRT tests). Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 3, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove orphaned waives whose only test entry is now gone: the GH200/sm100 video-neva platform waives, and the sm100 "Disable for Blackwell" combined-job waive (its multi-target l0_a100 tensorrt job was removed with the TRT tests). Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 4, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove orphaned waives whose only test entry is now gone: the GH200/sm100 video-neva platform waives, and the sm100 "Disable for Blackwell" combined-job waive (its multi-target l0_a100 tensorrt job was removed with the TRT tests). Also unschedule the legacy TensorRT-engine Triton tests in triton_server/test_triton.py (test_gpt, test_gptj, test_mistral*, test_whisper, test_medusa, test_eagle, test_llava*, test_mllama, test_qwen2_vl, test_t5_ib, test_gpt_ib*, etc.). Each builds a TRT engine via triton_server/build_model.sh, which pushd's into the examples/models/** dirs removed above and fails once they are gone ("pushd: examples/models/contrib/gptj: No such file or directory"). Drop their entries from l0_a30.yml, l0_a100.yml and l0_b200.yml (removing the now-empty a100/b200 triton post_merge blocks so trt-test-db does not choke on a null tests list) and their orphaned waives.txt lines. The self-contained Triton unit tests that build no engine are kept (test_cpp_unit_tests, test_python_bls_unit_tests, test_python_preproc_unit_tests, test_python_multimodal_encoders_unit_tests, test_fill_template, test_llmapi_unit_tests); PyTorch-path Triton coverage lives in test_triton_llmapi.py. Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 6, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove orphaned waives whose only test entry is now gone: the GH200/sm100 video-neva platform waives, and the sm100 "Disable for Blackwell" combined-job waive (its multi-target l0_a100 tensorrt job was removed with the TRT tests). Also unschedule the legacy TensorRT-engine Triton tests in triton_server/test_triton.py (test_gpt, test_gptj, test_mistral*, test_whisper, test_medusa, test_eagle, test_llava*, test_mllama, test_qwen2_vl, test_t5_ib, test_gpt_ib*, etc.). Each builds a TRT engine via triton_server/build_model.sh, which pushd's into the examples/models/** dirs removed above and fails once they are gone ("pushd: examples/models/contrib/gptj: No such file or directory"). Drop their entries from l0_a30.yml, l0_a100.yml and l0_b200.yml (removing the now-empty a100/b200 triton post_merge blocks so trt-test-db does not choke on a null tests list) and their orphaned waives.txt lines. The self-contained Triton unit tests that build no engine are kept (test_cpp_unit_tests, test_python_bls_unit_tests, test_python_preproc_unit_tests, test_python_multimodal_encoders_unit_tests, test_fill_template, test_llmapi_unit_tests); PyTorch-path Triton coverage lives in test_triton_llmapi.py. Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 6, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove orphaned waives whose only test entry is now gone: the GH200/sm100 video-neva platform waives, and the sm100 "Disable for Blackwell" combined-job waive (its multi-target l0_a100 tensorrt job was removed with the TRT tests). Also unschedule the legacy TensorRT-engine Triton tests in triton_server/test_triton.py (test_gpt, test_gptj, test_mistral*, test_whisper, test_medusa, test_eagle, test_llava*, test_mllama, test_qwen2_vl, test_t5_ib, test_gpt_ib*, etc.). Each builds a TRT engine via triton_server/build_model.sh, which pushd's into the examples/models/** dirs removed above and fails once they are gone ("pushd: examples/models/contrib/gptj: No such file or directory"). Drop their entries from l0_a30.yml, l0_a100.yml and l0_b200.yml (removing the now-empty a100/b200 triton post_merge blocks so trt-test-db does not choke on a null tests list) and their orphaned waives.txt lines. The self-contained Triton unit tests that build no engine are kept (test_cpp_unit_tests, test_python_bls_unit_tests, test_python_preproc_unit_tests, test_python_multimodal_encoders_unit_tests, test_fill_template, test_llmapi_unit_tests); PyTorch-path Triton coverage lives in test_triton_llmapi.py. Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
added a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 6, 2026
Follow-up to NVIDIA#15763 ([TRTLLM-13781] Remove legacy TensorRT examples), which deleted the TRT example scripts and waived their now-unrunnable tests with the note "tests to be removed in follow-up PR". This is that follow-up. Remove the waived test entries from the test-db/*.yml schedule lists and drop their matching SKIP lines from waives.txt. Covers examples/ (gpt, enc_dec, multimodal, whisper, redrafter, medusa, draft_target_model, eagle, commandr, llama, qwen, phi, mamba, granite, bert), accuracy/ test_cli_flow.py, and the TRT-engine cpp/ tests (test_e2e.py, test_multi_gpu.py executor/disagg cases). Also drop the now-empty 1-GPU tensorrt condition block in l0_a100.yml so trt-test-db does not choke on a null tests list, and remove orphaned waives whose only test entry is now gone: the GH200/sm100 video-neva platform waives, and the sm100 "Disable for Blackwell" combined-job waive (its multi-target l0_a100 tensorrt job was removed with the TRT tests). Also unschedule the legacy TensorRT-engine Triton tests in triton_server/test_triton.py (test_gpt, test_gptj, test_mistral*, test_whisper, test_medusa, test_eagle, test_llava*, test_mllama, test_qwen2_vl, test_t5_ib, test_gpt_ib*, etc.). Each builds a TRT engine via triton_server/build_model.sh, which pushd's into the examples/models/** dirs removed above and fails once they are gone ("pushd: examples/models/contrib/gptj: No such file or directory"). Drop their entries from l0_a30.yml, l0_a100.yml and l0_b200.yml (removing the now-empty a100/b200 triton post_merge blocks so trt-test-db does not choke on a null tests list) and their orphaned waives.txt lines. The self-contained Triton unit tests that build no engine are kept (test_cpp_unit_tests, test_python_bls_unit_tests, test_python_preproc_unit_tests, test_python_multimodal_encoders_unit_tests, test_fill_template, test_llmapi_unit_tests); PyTorch-path Triton coverage lives in test_triton_llmapi.py. Restore six shared multi-GPU C++ unit tests that NVIDIA#15763 over-waived by bulk-waiving all of cpp/test_multi_gpu.py: test_mpi_utils, test_fused_gemm_allreduce, three test_cache_transceiver params, and test_user_buffer. These run pure gtest binaries from tests/unit_tests/multi_gpu (mpiUtilsTest, gemmAllReduceTest, cacheTransceiverTest, userBufferTest) with synthetic inputs and need no example model or TRT engine; they were running before NVIDIA#15763. They are re-added as active in l0_dgx_h100.yml. test_cache_transceiver mooncake keeps its real nvbugs/5838199 waive. The test .py files are left in place — they still hold surviving PyTorch-path cases (e.g. test_gpt_oss_20b_lora_torch). The one whole-file waive (unittest/bindings/test_executor_bindings.py) is kept: every test there builds a TRT engine via trtllm.Executor, and its removal belongs with the C++ Executor-bindings removal, not this test-list cleanup. Scoped to tests/ — no runtime source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
This was referenced Jul 6, 2026
Closed
BrianLi23
pushed a commit
to BrianLi23/TensorRT-LLM
that referenced
this pull request
Jul 9, 2026
Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
This was referenced Jul 9, 2026
Merged
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Summary
This is step 1.1 of removing the legacy TensorRT backend from TensorRT-LLM.
It removes the per-model TensorRT example directories (the
convert_checkpoint.py→trtllm-build→ enginerun.pyworkflow) whilekeeping all PyTorch-backend example content, and waives the legacy-TensorRT
tests that build engines from those now-removed examples so CI stays green.
PyTorch is the supported backend going forward; it loads HuggingFace checkpoints
directly, so the per-model checkpoint-conversion / engine-build examples no longer
apply.
Scope:
examples/(deletions) +pyproject.toml(one codespell cleanup) +tests/integration/test_lists/waives.txt(waive the dependent legacy-TRT tests).154 files changed, +65 / −34,623— zero Python/C++ source touched, so this PRcannot change runtime behavior or break the package build /
import tensorrt_llm.This is one of four directory-scoped PRs that split the TensorRT-example/doc/test
removal for easier review (1.1 examples, 1.2 docs, 1.3 tests, 1.4 Triton C++ backend).
What is removed (149 deletions)
The TensorRT example trees — each is the
convert_checkpoint.py/trtllm-buildengine workflow:
examples/models/core/*— bert, commandr, enc_dec, gemma, glm-4-9b, gpt,internlm2, llama, mamba, mllama, nemotron_nas, phi, qwen, recurrentgemma, vit, whisper
examples/models/contrib/*— baichuan, bloom, dbrx, deepseek_v1/v2, dit, mmdit,stdit, falcon, gptj, gptneox, grok, mpt, opt, cogvlm, …
examples/eagle/,examples/medusa/,examples/redrafter/,examples/lookahead/,examples/draft_target_model/examples/llm-api/_tensorrt_engine/(the TensorRT-engineLLMmirror)What is kept (PyTorch content preserved)
examples/models/core/phi/phi4-mm.mdexamples/llm-api/quickstart_multimodal.py,--backend pytorch, referencestensorrt_llm/_torch/models/modeling_phi4mm.pyexamples/models/core/qwen/README.mdquickstart_advanced.py,trtllm-eval/trtllm-bench --backend pytorch,trtllm-serve); dropped the TRT Qwen/Qwen2 convert/build bodyexamples/models/core/gemma/README.mdtrtllm-serve,trtllm-eval); dropped the older Gemma TRT build flowexamples/models/core/nemotron_nas/README.mdexamples/llm-api/quickstart_advanced.py/trtllm-serveNot touched at all:
examples/llm-api/,examples/auto_deploy/,examples/serve/,examples/models/core/multimodal/, and the rest of the non-TRT examples.Other source change
pyproject.toml: drop thecodespellexclude-fileentry that pointed at theremoved
examples/models/core/whisper/tokenizer.py.CI: waive the legacy-TRT tests that depend on the removed examples
The removed example directories are not merely user docs — a body of legacy
TensorRT tests builds engines directly from them at test time (via the
convert_checkpoint.py→trtllm-buildworkflow,*_example_rootconftestfixtures, the
accuracy_coreEXAMPLE_FOLDERpath, and the C++cpp/tests/resources/scripts/build_*_engines.pyscripts). With the examples gonethese tests fail with missing-file errors (
convert_checkpoint.py/requirements.txt/ generatedconfig.json).Because these are all legacy TensorRT-backend tests that will be removed
together with the backend, we waive them here (rather than restoring the examples)
and remove both the tests and these waives in a follow-up PR. Every waive is
tagged
TRTLLM-13781so the follow-up cangrepand delete them as a unit.41 waive entries added to
tests/integration/test_lists/waives.txt:cpp/test_e2e.py::test_model[-eagle-86 / -medusa-86 / -recurrentgemma-86 / -gpt_executor-80 / -gpt_tests-80 / -t5-90 / -enc_dec_language_adapter-90 / fp8-llama-90],::test_benchmarks[t5-90]cpp/test_multi_gpu.pyunittest/bindings/test_executor_bindings.py,unittest/test_model_runner_cpp.pyaccuracy/test_cli_flow.pyEXAMPLE_FOLDER→convert_checkpoint+trtllm-build)examples/test_{bert,whisper,enc_dec,medusa,redrafter,eagle,draft_target_model,mamba,commandr,nemotron_nas,qwen2audio,qwen,multimodal}.pytest_gpt.py(gpt2 medium / bad_words / stop_words / next_prompt_tuning),test_llama.py(long_alpaca_8gpu_summary, v3_dora_1gpu),test_phi.py(phi_quantization_1gpu),test_granite.py(granite_bf16_lora, llm_granite)Scoping & safety
only to files verified to contain no PyTorch/AutoDeploy tests. For the mixed
files (
test_gpt/test_llama/test_phi/test_granite) only the individualTensorRT test functions are waived (via the exact parametrized node-ids
present in the active test lists); the PyTorch tests in the same files —
test_gpt_oss_20b_lora_torch,test_llama_3_x_with_bf16_lora_torch,test_phi_4_mini_instruct_with_bf16_lora_torch— keep running.use_py_session/use_cpp_sessionparametrizations are the TensorRT Python /C++ runtimes (still TRT engines), not the PyTorch backend, and are correctly waived.
test_benchmarks[bart-90]/[gpt-80]and threetest_graniteparams were already waived under existing NVBugs and were left untouched.
Validation of the waive list
scripts/check_test_list.py --check-duplicate-waives→ pass (no duplicates).scripts/check_test_list.py --validate→ pass (AST: 2695 unique test entries; everywaive resolves to a real test id present in the active test lists).
waives.txtstaysLC_ALL=C-sorted; passes the test-list txt tab/space andfile-contents-sorter hooks.
How files were classified
Every deleted file was read from its original content and classified
TRT-ONLY / KEEP-PYTORCH / MIXED using concrete signals:
convert_checkpoint.py/
trtllm-buildusage,--backend/backend=values,_tensorrt_engineimports, and references to
tensorrt_llm/_torch/. All 117 deleted scripts and 31deleted READMEs are TRT-only; the 4 files above carried PyTorch content and were
preserved.
Validation
tensorrt_llm/**orcpp/**changed → no build/import/runtime impact.(
examples/llm-api/quickstart_*.py) resolve.test_model[-eagle/-medusa],A10-TensorRT
test_executor_bindings/test_bert/test_whisper, and theH100 cpp /
test_cli_flow/test_enc_dec/test_model_runner_cppfailures)map to a waive entry and are skipped.
Summary by CodeRabbit
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
If PR introduces API changes, an appropriate PR label is added - either
api-compatibleorapi-breaking. Forapi-breaking, includeBREAKINGin the PR title.Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.