[ML] Harden pytorch_inference with TorchScript model graph validation by edsavage · Pull Request #2999 · elastic/ml-cpp

edsavage · 2026-03-15T21:16:07Z

Summary

Re-applies #2936 and #2991 which were reverted in #2995.

Adds a static TorchScript graph validation layer (CModelGraphValidator, CSupportedOperations) that rejects models containing operations not observed in supported transformer architectures, reducing the attack surface by ensuring only known-safe operation sets are permitted.
Includes aten::mul_ and quantized::linear_dynamic in the allowed operations for dynamically quantized models (e.g. ELSER v2 imported via Eland).
Adds Python extraction tooling (dev-tools/extract_model_ops/) to trace reference HuggingFace models and collect their op sets, with support for quantized variants.
Adds reference_model_ops.json golden file and C++ drift test to detect allowlist staleness on PyTorch upgrades.
Adds adversarial "evil model" integration tests to verify rejection of forbidden operations.
Adds CHANGELOG entry.

Test plan

Local test_pytorch_inference passes
CI builds pass on all platforms
QA tests pass (ci:run-qa-tests label applied) See https://buildkite.com/elastic/appex-qa-stateful-custom-ml-cpp-build-testing/builds/724

This reverts commit 4f1ec3e.

prodsecmachine · 2026-03-15T21:16:24Z

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.

…ns quotes Sanitize BUILDKITE_MESSAGE before embedding in generated pipeline YAML to prevent double quotes and multi-line content from breaking the YAML structure. Affects both run_qa_tests.yml.sh and run_pytorch_tests.yml.sh. Made-with: Cursor

Add aten::bmm, aten::ceil, aten::floor_divide, aten::gt, aten::le, and aten::sign to the allowed operations list, fixing graph validation failures for the .rerank-v1 model used by the default rerank endpoint. These operations were identified by running the full Elasticsearch inference and ML integration test suites (974 tests) against a build with graph validation enabled. Made-with: Cursor

Add aten::clamp_min, aten::eq, aten::expand_as, aten::linalg_vector_norm, and aten::sum to the allowed operations list, fixing graph validation failures for distilbert-base-uncased-finetuned-sst-2-english and sentence-transformers/all-distilroberta-v1 models. Made-with: Cursor

…d additional model ops Add .rerank-v1 (52 ops extracted from ml-models.elastic.co .pt file), distilbert-sst2, all-distilroberta-v1, and their Eland-deployed variants to the reference model golden file. All ops extracted with PyTorch 2.7.1. Add aten::detach (Eland-traced models) and aten::masked_fill_ (.rerank-v1) to the allowlist. Made-with: Cursor

Add a parallel Buildkite step that runs Elasticsearch inference integration tests against the ml-cpp build artifacts. The new step runs on its own machine alongside the existing ES tests step. Tests exercised: - DefaultEndPointsIT (ELSER, E5, rerank default endpoints) - TextEmbeddingCrudIT (E5 model CRUD via inference API) - Semantic text YAML REST tests (indexing and querying with default ELSER 2 endpoint) All tests use local prepacked models served by the test framework — no external services required. Made-with: Cursor

Extract clone/branch selection, Java configuration, and Gradle invocation into run_es_tests_common.sh. Both run_es_tests.sh and run_es_inference_tests.sh are now thin wrappers that pass their Gradle commands as arguments to the common script. Made-with: Cursor

Resolve conflict in dev-tools/run_es_tests.sh: keep the refactored thin-wrapper structure from this branch and integrate the Gradle build cache support (from elastic#2907) into run_es_tests_common.sh. Made-with: Cursor

Copilot

Pull request overview

Reintroduces and extends TorchScript model-graph validation for pytorch_inference to reduce attack surface by enforcing an operation allowlist/denylist, and adds tooling + tests to keep that allowlist current (including quantized model support).

Changes:

Add C++ TorchScript graph validation (CModelGraphValidator + CSupportedOperations) and wire it into pytorch_inference startup.
Add Python tooling (dev-tools/extract_model_ops/) + CMake runner to extract/validate op allowlists and detect drift via a golden JSON + C++ test.
Add adversarial TorchScript fixtures and integration tests; extend Buildkite scripts to run additional Elasticsearch inference integration tests.

Reviewed changes

Copilot reviewed 22 out of 45 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
`test/CMakeLists.txt`	Runs allowlist validation script as part of `test_all_parallel` (optional) and adds a standalone validation target.
`docs/CHANGELOG.asciidoc`	Adds release note entry for graph validation hardening.
`dev-tools/run_es_tests_common.sh`	Factors shared logic for running ES integration tests from a local Ivy repo.
`dev-tools/run_es_tests.sh`	Uses the common runner to execute core ML REST/YAML tests.
`dev-tools/run_es_inference_tests.sh`	Adds separate runner for ES inference integration tests.
`dev-tools/generate_malicious_models.py`	Adds generator for malicious TorchScript fixtures used by validator integration tests.
`dev-tools/extract_model_ops/validation_models.json`	Defines HF model set to validate allowlist (incl. quantized variants).
`dev-tools/extract_model_ops/validate_allowlist.py`	Adds Python-side allowlist validation against traced models + local `.pt` fixtures.
`dev-tools/extract_model_ops/torchscript_utils.py`	Shared tracing/inlining + config loading helpers (incl. dynamic quantization).
`dev-tools/extract_model_ops/requirements.txt`	Pins Python deps for extraction/validation tooling (torch + transformers stack).
`dev-tools/extract_model_ops/reference_models.json`	Defines reference HF models used to build the allowlist union (incl. quantized).
`dev-tools/extract_model_ops/extract_model_ops.py`	Adds extractor to generate op unions / C++ initializer / golden JSON.
`dev-tools/extract_model_ops/es_it_models/tiny_text_expansion.pt`	Adds ES IT TorchScript fixture for validation.
`dev-tools/extract_model_ops/es_it_models/tiny_text_embedding.pt`	Adds ES IT TorchScript fixture for validation.
`dev-tools/extract_model_ops/es_it_models/supersimple_pytorch_model_it.pt`	Adds ES IT TorchScript fixture for validation.
`dev-tools/extract_model_ops/es_it_models/README.md`	Documents provenance/regeneration of ES IT TorchScript fixtures.
`dev-tools/extract_model_ops/README.md`	Documents extraction/validation workflows and golden drift test.
`dev-tools/extract_model_ops/.gitignore`	Ignores the local Python venv used by the tooling.
`cmake/run-validation.cmake`	Adds portable CMake driver to create venv, install deps, and run validation.
`cmake/functions.cmake`	Wires validation into `precommit` target (optional).
`bin/pytorch_inference/unittest/testfiles/reference_model_ops.json`	Adds/updates golden per-model op sets for drift detection tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_rop_exploit.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_mixed_file_reader.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_many_unrecognised.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_hidden_in_submodule.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_heap_leak.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_file_reader_in_submodule.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_file_reader.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_conditional.pt`	Adds malicious TorchScript fixture used by integration tests.
`bin/pytorch_inference/unittest/CThreadSettingsTest.cc`	Switches includes to angle-bracket form.
`bin/pytorch_inference/unittest/CResultWriterTest.cc`	Switches includes to angle-bracket form.
`bin/pytorch_inference/unittest/CModelGraphValidatorTest.cc`	Adds unit + integration tests for validator, fixtures, and allowlist drift.
`bin/pytorch_inference/unittest/CMakeLists.txt`	Adds validator test source and include path for new include style.
`bin/pytorch_inference/unittest/CCommandParserTest.cc`	Switches includes to angle-bracket form.
`bin/pytorch_inference/Main.cc`	Enforces model-graph validation at load time; improves rejection messages.
`bin/pytorch_inference/CSupportedOperations.h`	Declares forbidden/allowed op sets for validation.
`bin/pytorch_inference/CSupportedOperations.cc`	Defines forbidden/allowed TorchScript ops (incl. quantized ops).
`bin/pytorch_inference/CModelGraphValidator.h`	Declares validator API + node-count guard.
`bin/pytorch_inference/CModelGraphValidator.cc`	Implements graph inlining + op collection + allow/deny evaluation.
`bin/pytorch_inference/CMakeLists.txt`	Builds new validator + supported-ops sources into `pytorch_inference`.
`.buildkite/scripts/steps/run_es_inference_tests.sh`	Adds Buildkite step script for ES inference integration tests.
`.buildkite/pipelines/run_qa_tests.yml.sh`	Sanitizes Buildkite message when triggering downstream QA pipeline.
`.buildkite/pipelines/run_pytorch_tests.yml.sh`	Sanitizes Buildkite message when triggering downstream PyTorch pipeline.
`.buildkite/pipelines/run_es_inference_tests_x86_64.yml.sh`	Adds x86_64 pipeline to run ES inference integration tests.
`.buildkite/pipeline.json.py`	Uploads the new ES inference tests runner pipeline when x86_64 enabled.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

dev-tools/extract_model_ops/validate_allowlist.py

cmake/run-validation.cmake

.buildkite/pipelines/run_pytorch_tests.yml.sh

cmake/functions.cmake

test/CMakeLists.txt

cmake/run-validation.cmake

dev-tools/run_es_tests_common.sh

.buildkite/pipelines/run_qa_tests.yml.sh

- Remove unused collect_graph_ops import and fix help text in validate_allowlist.py - Query venv Python for site-packages path instead of globbing (which can yield multiple paths) in run-validation.cmake - Bump cmake_minimum_required to 3.19.2 to match the repo - Escape backslashes in SAFE_MESSAGE for YAML double-quoted strings in pipeline scripts - Remove allowlist validation from precommit and test_all_parallel targets to keep them fast; validation remains available via the standalone validate_pytorch_inference_models target - Document why eval is necessary in run_es_tests_common.sh and properly quote the Ivy repo URL Made-with: Cursor

Resolve conflict in dev-tools/run_es_tests.sh: incorporate ES_TEST_SUITE support from elastic#2990 (parallel javaRestTest/yamlRestTest steps) into our thin-wrapper architecture that delegates to run_es_tests_common.sh. Made-with: Cursor

Copilot

Pull request overview

Re-introduces TorchScript model graph validation for pytorch_inference to reduce attack surface by rejecting models that contain forbidden or unrecognised TorchScript ops, and adds tooling/tests to keep the allowlist in sync with supported model architectures (including quantized variants).

Changes:

Added C++ TorchScript graph validator + op allowlist/denylist, and wired validation into pytorch_inference startup.
Added C++ unit/integration tests, golden op-set drift test, and malicious-model fixtures to verify forbidden/unrecognised ops are rejected.
Added developer tooling + CMake runner to extract/validate op allowlists against reference HF models and local ES IT .pt models; added Buildkite scripts to run ES inference ITs.

Reviewed changes

Copilot reviewed 21 out of 44 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
`test/CMakeLists.txt`	Adds a standalone CMake target to run the Python allowlist validation script.
`docs/CHANGELOG.asciidoc`	Documents the TorchScript graph validation hardening enhancement.
`dev-tools/run_es_tests_common.sh`	Factors out shared ES integration test runner logic (clone/setup/gradle/cache).
`dev-tools/run_es_tests.sh`	Delegates ES ML REST/YAML test execution to the common runner.
`dev-tools/run_es_inference_tests.sh`	Adds a dedicated runner for ES inference integration test suites.
`dev-tools/generate_malicious_models.py`	Adds generator for malicious TorchScript model fixtures used by validator tests.
`dev-tools/extract_model_ops/validation_models.json`	Defines HF models used by the Python allowlist validation run.
`dev-tools/extract_model_ops/validate_allowlist.py`	Implements Python-side allowlist validation against C++ allow/deny sets.
`dev-tools/extract_model_ops/torchscript_utils.py`	Shared utilities for tracing/scripting HF models and collecting inlined ops.
`dev-tools/extract_model_ops/requirements.txt`	Pins Python dependencies for extraction/validation tooling.
`dev-tools/extract_model_ops/reference_models.json`	Defines HF reference models used to derive the allowlist/golden file.
`dev-tools/extract_model_ops/extract_model_ops.py`	Implements op extraction and golden file generation for allowlist maintenance.
`dev-tools/extract_model_ops/es_it_models/tiny_text_expansion.pt`	Adds pre-saved ES IT TorchScript model fixture for validation.
`dev-tools/extract_model_ops/es_it_models/tiny_text_embedding.pt`	Adds pre-saved ES IT TorchScript model fixture for validation.
`dev-tools/extract_model_ops/es_it_models/supersimple_pytorch_model_it.pt`	Adds pre-saved ES IT TorchScript model fixture for validation.
`dev-tools/extract_model_ops/es_it_models/README.md`	Documents provenance/regeneration steps for ES IT `.pt` fixtures.
`dev-tools/extract_model_ops/README.md`	Documents how to extract/validate ops and regenerate golden drift file.
`dev-tools/extract_model_ops/.gitignore`	Ignores the local venv used by extraction tooling.
`cmake/run-validation.cmake`	Adds portable CMake script to create venv, install deps, and run validation.
`bin/pytorch_inference/unittest/testfiles/reference_model_ops.json`	Adds/updates golden file for allowlist drift detection in C++ tests.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_rop_exploit.pt`	Adds malicious fixture to ensure validator rejects as_strided-based exploit graphs.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_mixed_file_reader.pt`	Adds malicious fixture to ensure validator rejects forbidden file access ops.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_many_unrecognised.pt`	Adds malicious fixture to ensure validator rejects unexpected/unrecognised ops.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_hidden_in_submodule.pt`	Adds malicious fixture to ensure validator detects ops hidden in submodules.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_heap_leak.pt`	Adds malicious fixture for heap-leak style graphs to validate rejection.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_file_reader_in_submodule.pt`	Adds malicious fixture for forbidden ops inside submodules.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_file_reader.pt`	Adds malicious fixture to ensure validator rejects `aten::from_file`.
`bin/pytorch_inference/unittest/testfiles/malicious_models/malicious_conditional.pt`	Adds malicious fixture to ensure validator scans conditional branches.
`bin/pytorch_inference/unittest/CThreadSettingsTest.cc`	Switches unittest includes to use configured include dirs (`<...>`).
`bin/pytorch_inference/unittest/CResultWriterTest.cc`	Switches unittest includes to use configured include dirs (`<...>`).
`bin/pytorch_inference/unittest/CModelGraphValidatorTest.cc`	Adds comprehensive unit/integration tests for graph validation and drift checks.
`bin/pytorch_inference/unittest/CMakeLists.txt`	Wires new validator tests and adds include path for `<...>` style includes.
`bin/pytorch_inference/unittest/CCommandParserTest.cc`	Switches unittest includes to use configured include dirs (`<...>`).
`bin/pytorch_inference/Main.cc`	Calls `CModelGraphValidator` on model load and fails fast on invalid graphs.
`bin/pytorch_inference/CSupportedOperations.h`	Declares the allowed/forbidden operation sets for graph validation.
`bin/pytorch_inference/CSupportedOperations.cc`	Defines the allowed/forbidden operation sets (incl. quantized ops).
`bin/pytorch_inference/CModelGraphValidator.h`	Declares validator API and collection logic for inlined TorchScript graphs.
`bin/pytorch_inference/CModelGraphValidator.cc`	Implements graph inlining, op collection, and allow/deny validation logic.
`bin/pytorch_inference/CMakeLists.txt`	Adds validator and allowlist sources to the `pytorch_inference` build.
`.buildkite/scripts/steps/run_es_inference_tests.sh`	Adds Buildkite step script to run ES inference integration tests.
`.buildkite/pipelines/run_qa_tests.yml.sh`	Sanitizes Buildkite message when triggering downstream QA pipeline.
`.buildkite/pipelines/run_pytorch_tests.yml.sh`	Sanitizes Buildkite message when triggering downstream PyTorch test pipeline.
`.buildkite/pipelines/run_es_inference_tests_x86_64.yml.sh`	Adds Buildkite pipeline generator for x86_64 inference ITs.
`.buildkite/pipeline.json.py`	Wires the new inference IT runner pipeline upload into the main pipeline.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

dev-tools/run_es_tests_common.sh

+for GRADLE_CMD in "$@" ; do
+    # eval is required here because each GRADLE_CMD argument from the caller
+    # contains embedded shell quoting (e.g. --tests "class.name {p0=glob/*}")
+    # that must be interpreted by the shell.  The environment variables
+    # (GRADLE_JVM_OPTS, CACHE_ARGS, EXTRA_TEST_OPTS) are set by our own
+    # scripts and the CI environment, not by untrusted input.
+    eval ./gradlew $GRADLE_JVM_OPTS $CACHE_ARGS \
+        "-Dbuild.ml_cpp.repo=$IVY_REPO_URL" \
+        $GRADLE_CMD $EXTRA_TEST_OPTS


cmake/run-validation.cmake

+find_program(_python_path
+  NAMES python3 python3.12 python3.11 python3.10 python3.9 python
+  DOC "Python 3 interpreter"


dev-tools/extract_model_ops/validate_allowlist.py

+def load_pt_and_collect_ops(pt_path: str) -> set[str] | None:
+    """Load a saved TorchScript .pt file, inline, and return its op set."""
+    try:
+        module = torch.jit.load(pt_path)
+        return collect_inlined_ops(module)
+    except Exception as exc:
+        print(f"    LOAD ERROR: {exc}", file=sys.stderr)
+        return None


dev-tools/extract_model_ops/extract_model_ops.py

+def extract_ops_for_model(model_name: str,
+                          quantize: bool = False) -> set[str] | None:
+    """Trace a HuggingFace model and return its TorchScript op set.
+
+    Returns None if the model could not be loaded or traced.
+    """
+    label = f"{model_name} (quantized)" if quantize else model_name
+    print(f"  Loading {label}...", file=sys.stderr)


edsavage · 2026-03-17T23:57:53Z

@copilot open a new pull request to apply changes based on the comments in this thread

- Replace PEP 604 union types (set[str] | None) with Optional[set[str]] for Python 3.9 compatibility - Drop python3.9 from CMake find_program search list since scripts use PEP 585 generics requiring 3.9+ - Eliminate eval in run_es_tests_common.sh by changing the caller interface: each Gradle argument is now a separate shell argument with '---' as a sentinel separating multiple invocations - Convert CACHE_ARGS from string to proper bash array Made-with: Cursor

edsavage · 2026-03-19T00:37:30Z

buildkite run_qa_tests

valeriy42

LGTM

valeriy42 · 2026-03-19T13:48:32Z

.buildkite/pipelines/run_es_inference_tests_x86_64.yml.sh

Great that you are fixing this test gap!

…lidation (elastic#2999)" This reverts commit d3df09c.

…lidation (#2999)" (#3006) This reverts commit d3df09c.

…alidation (elastic#2999)" (elastic#3006) This reverts commit ceabc9b.

…alidation (#3008) Reapply "[ML] Harden pytorch_inference with TorchScript model graph validation (#2999)" (#3006) This reverts commit ceabc9b. - Adds a static TorchScript graph validation layer (CModelGraphValidator, CSupportedOperations) that rejects models containing operations not observed in supported transformer architectures, reducing the attack surface by ensuring only known-safe operation sets are permitted. - Includes aten::mul_ and quantized::linear_dynamic in the allowed operations for dynamically quantized models (e.g. ELSER v2 imported via Eland). - Adds Python extraction tooling (dev-tools/extract_model_ops/) to trace reference HuggingFace models and collect their op sets, with support for quantized variants. - Adds reference_model_ops.json golden file and C++ drift test to detect allowlist staleness on PyTorch upgrades. - Adds adversarial "evil model" integration tests to verify rejection of forbidden operations. - Adds CHANGELOG entry. - Add aten::norm to graph validator allowlist The prepacked .multilingual-e5-small model uses aten::norm for normalization, which was not in the allowlist. This caused the model to be rejected with "Unrecognised operations: aten::norm". - Add multilingual-e5-small model ops to reference files Extracted ops from intfloat/multilingual-e5-small (base and Eland text_embedding variant) and added both to the reference golden file. The base model uses standard XLM-RoBERTa ops. The Eland variant adds pooling/normalization ops (linalg_vector_norm, clamp, etc.). The prepacked .multilingual-e5-small model bundled with Elasticsearch uses aten::norm (added to the allowlist in the previous commit). - Add graph validator test for prepacked e5 model with aten::norm The prepacked .multilingual-e5-small model uses aten::norm, which was missing from the allowlist and caused production failures. This test loads a tiny (24KB) model that mirrors the real prepacked model's graph structure (including aten::norm) and verifies graph validation passes. The test model was created by tracing a minimal XLM-RoBERTa-like architecture with normalization, then patching the TorchScript IR to use aten::norm (which modern PyTorch decomposes into aten::linalg_vector_norm, so it can't be generated via tracing). Made-with: Cursor

Revert "[ML] Revert elastic#2991 and elastic#2936 (elastic#2995)"

efb9d68

This reverts commit 4f1ec3e.

edsavage added >enhancement :ml ci:run-qa-tests Run a subset of the QA tests v9.4.0 labels Mar 15, 2026

edsavage marked this pull request as draft March 15, 2026 21:19

edsavage added 6 commits March 16, 2026 11:18

edsavage requested a review from Copilot March 17, 2026 00:43

Copilot started reviewing on behalf of edsavage March 17, 2026 00:44 View session

Merge upstream/main into feature/harden-pytorch-inference-v2

e84563a

Resolve conflict in dev-tools/run_es_tests.sh: keep the refactored thin-wrapper structure from this branch and integrate the Gradle build cache support (from elastic#2907) into run_es_tests_common.sh. Made-with: Cursor

Copilot AI reviewed Mar 17, 2026

View reviewed changes

edsavage added 2 commits March 17, 2026 14:32

edsavage marked this pull request as ready for review March 17, 2026 22:50

edsavage requested review from Copilot and valeriy42 March 17, 2026 22:50

Copilot started reviewing on behalf of edsavage March 17, 2026 22:50 View session

Copilot AI reviewed Mar 17, 2026

View reviewed changes

valeriy42 approved these changes Mar 19, 2026

View reviewed changes

.buildkite/pipelines/run_es_inference_tests_x86_64.yml.sh

Copy link

Contributor

valeriy42 Mar 19, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great that you are fixing this test gap!

edsavage merged commit d3df09c into elastic:main Mar 19, 2026
20 checks passed

edsavage added a commit to edsavage/ml-cpp that referenced this pull request Mar 20, 2026

Revert "[ML] Harden pytorch_inference with TorchScript model graph va…

0caa2bf

…lidation (elastic#2999)" This reverts commit d3df09c.

edsavage mentioned this pull request Mar 20, 2026

[ML] Revert: Harden pytorch_inference with TorchScript model graph validation #3006

Merged

edsavage added a commit that referenced this pull request Mar 20, 2026

Revert "[ML] Harden pytorch_inference with TorchScript model graph va…

ceabc9b

…lidation (#2999)" (#3006) This reverts commit d3df09c.

edsavage added a commit to edsavage/ml-cpp that referenced this pull request Mar 22, 2026

Reapply "[ML] Harden pytorch_inference with TorchScript model graph v…

e6e5fc5

…alidation (elastic#2999)" (elastic#3006) This reverts commit ceabc9b.

edsavage mentioned this pull request Mar 22, 2026

[ML] Reapply: Harden pytorch_inference with TorchScript model graph validation #3008

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[ML] Harden pytorch_inference with TorchScript model graph validation#2999

[ML] Harden pytorch_inference with TorchScript model graph validation#2999
edsavage merged 11 commits intoelastic:mainfrom
edsavage:feature/harden-pytorch-inference-v2

edsavage commented Mar 15, 2026 •

edited

Loading

Uh oh!

prodsecmachine commented Mar 15, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

edsavage commented Mar 17, 2026

Uh oh!

edsavage commented Mar 19, 2026

Uh oh!

valeriy42 left a comment

Uh oh!

valeriy42 Mar 19, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

edsavage commented Mar 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

prodsecmachine commented Mar 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Snyk checks have passed. No issues have been found so far.

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

edsavage commented Mar 17, 2026

Uh oh!

edsavage commented Mar 19, 2026

Uh oh!

valeriy42 left a comment

Choose a reason for hiding this comment

Uh oh!

valeriy42 Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

edsavage commented Mar 15, 2026 •

edited

Loading

prodsecmachine commented Mar 15, 2026 •

edited

Loading