Merge branch 'main' into export-D103467782

Author: rascani

.ci/scripts/unittest-linux-cmake.sh

@@ -11,19 +11,18 @@ set -eux
 # Install from a local tosa-tools checkout when available. If absent in this
 # checkout layout, clone the pinned upstream tag and install from there.
 if ! python -c "import tosa_serializer" >/dev/null 2>&1; then
-  TOSA_SERIALIZATION_DIR="./examples/arm/arm-scratch/tosa-tools/serialization"
-  if [[ ! -d "${TOSA_SERIALIZATION_DIR}" ]]; then
+  TOSA_TOOLS_DIR="./examples/arm/arm-scratch/tosa-tools"
+  if [[ ! -d "${TOSA_TOOLS_DIR}" ]]; then
     TOSA_TOOLS_DIR="$(mktemp -d /tmp/tosa-tools.XXXXXX)"
-    git clone --depth 1 --branch v2025.11.2 \
+    git clone --depth 1 --branch v2026.05.0 \
       https://git.gitlab.arm.com/tosa/tosa-tools.git "${TOSA_TOOLS_DIR}"
-    TOSA_SERIALIZATION_DIR="${TOSA_TOOLS_DIR}/serialization"
   fi
 
   # NOTE: Will be removed when tosa-tools is installed via pypi
   python -m pip install pybind11==2.10.4
   CMAKE_POLICY_VERSION_MINIMUM=3.5 BUILD_PYBIND=1 \
     python -m pip install --no-dependencies \
-    "${TOSA_SERIALIZATION_DIR}"
+    "${TOSA_TOOLS_DIR}"
   python -c "import tosa_serializer"
 fi
 

.claude/skills/building/SKILL.md

@@ -23,17 +23,16 @@ conda activate executorch
 
 **Path B — no conda (fall back to venv):**
 ```bash
-# Find a compatible Python (3.10–3.13). On macOS with only Homebrew Python 3.14+,
-# install a compatible version first: brew install python@3.12
-python3.12 -m venv .executorch-venv   # or python3.11, python3.10, python3.13
+# Find a compatible Python (3.10–3.14).
+python3.12 -m venv .executorch-venv   # or python3.11, python3.10, python3.13, python3.14
 source .executorch-venv/bin/activate
 pip install --upgrade pip
 ```
 
 **Then verify (either path):**
 
 Run `python --version` and `cmake --version`. Fix automatically:
-- **Python not 3.10–3.13**: recreate the env with a correct Python version.
+- **Python not 3.10–3.14**: recreate the env with a correct Python version.
 - **cmake missing or < 3.24**: run `pip install 'cmake>=3.24'` inside the env.
 - **cmake >= 4.0**: works in practice, no action needed.
 

.github/workflows/add-unanswered-to-project.yml

@@ -63,7 +63,7 @@ jobs:
 
             // List of organization logins (lowercased) to exclude members of
             const excludedOrgs = new Set([
-              "meta", "facebook", "pytorch", "arm", "apple", "qualcomm", "nxp", "mediatek", "cadence", "intel", "samsung",
+              "meta", "facebook", "pytorch", "arm", "apple", "qualcomm", "nxp", "mediatek", "cadence", "intel", "samsung", "arm ltd.",
               "@meta", "@facebook", "@pytorch", "@arm", "@apple", "@qualcomm", "@nxp", "@mediatek", "@cadence", "@intel", "@samsung"
             ]);
 

.github/workflows/build-wheels-aarch64-linux.yml

@@ -30,7 +30,7 @@ jobs:
       test-infra-ref: main
       with-cuda: disabled
       with-rocm: disabled
-      python-versions: '["3.10", "3.11", "3.12", "3.13"]'
+      python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
 
   build:
     needs: generate-matrix

.github/workflows/build-wheels-linux.yml

@@ -30,7 +30,7 @@ jobs:
       test-infra-ref: main
       with-cuda: disabled
       with-rocm: disabled
-      python-versions: '["3.10", "3.11", "3.12", "3.13"]'
+      python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
 
   build:
     needs: generate-matrix

.github/workflows/build-wheels-macos.yml

@@ -30,7 +30,7 @@ jobs:
       test-infra-ref: main
       with-cuda: disabled
       with-rocm: disabled
-      python-versions: '["3.10", "3.11", "3.12", "3.13"]'
+      python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
 
   build:
     needs: generate-matrix

.github/workflows/build-wheels-windows.yml

@@ -33,7 +33,7 @@ jobs:
       test-infra-ref: main
       with-cuda: disabled
       with-rocm: disabled
-      python-versions: '["3.10", "3.11", "3.12", "3.13"]'
+      python-versions: '["3.10", "3.11", "3.12", "3.13", "3.14"]'
 
   build:
     needs: generate-matrix

.github/workflows/pull.yml

@@ -852,8 +852,7 @@ jobs:
     strategy:
       matrix:
         dtype: [fp32]
-        # TODO(T12345): re-enable qnn_16a16w once OOM on linux.2xlarge is resolved
-        pt2e_quantize: [qnn_8a8w]
+        pt2e_quantize: [qnn_16a16w, qnn_8a8w]
         mode: [qnn]
       fail-fast: false
     with:

.github/workflows/trunk.yml

@@ -951,8 +951,7 @@ jobs:
     strategy:
       matrix:
         dtype: [fp32]
-        # TODO(T12345): re-enable qnn_16a16w once OOM on linux.2xlarge is resolved
-        pt2e_quantize: [qnn_8a8w]
+        pt2e_quantize: [qnn_16a16w, qnn_8a8w]
         mode: [qnn]
       fail-fast: false
     with:

backends/arm/test/models/Qwen3_VL/test_qwen3_vl_model.py

@@ -252,16 +252,17 @@ def _test_qwen3_vl_full_models_tosa_FP(
 def _test_qwen3_vl_full_models_tosa_FP_bf16(
     test_case: Qwen3VLModelTestCase,
     config_factory=_make_qwen3_vl_e2e_test_config,
+    text_model_atol: float | None = None,
 ):
     model, inputs = test_case.model_cls.prepare_model_and_inputs(config_factory)
     model, inputs = _to_bfloat16_model_and_floating_inputs(model, inputs)
     # Slightly higher atol for TOSA BF16 on aarch64 (MLETORCH-2048: numeric mismatch)
-    atol = (
-        0.4
-        if common.is_aarch64_host()
-        and test_case.model_cls is LowerableVisionModelWrapper
-        else 0.1
-    )
+    if common.is_aarch64_host() and test_case.model_cls is LowerableVisionModelWrapper:
+        atol = 0.4
+    elif text_model_atol is not None and test_case.model_cls is TextModelWrapper:
+        atol = text_model_atol
+    else:
+        atol = 0.1
     with torch.no_grad():
         pipeline = TosaPipelineFP[input_t](
             model,
@@ -382,7 +383,9 @@ def test_qwen3_vl_2b_instruct_full_models_tosa_FP_bf16(
     test_case: Qwen3VLModelTestCase,
 ):
     _test_qwen3_vl_full_models_tosa_FP_bf16(
-        test_case, _make_qwen3_vl_2b_instruct_layer_config
+        test_case,
+        _make_qwen3_vl_2b_instruct_layer_config,
+        text_model_atol=0.15,
     )