compressed-tensors 0.10.3a20250620__tar.gz → 0.10.3a20250703__tar.gz

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/.github/actions/test/action.yml

@@ -22,6 +22,23 @@ runs:
           name: compressed
           extra: "[dev,accelerate]"
 
+    - name: clean up
+      run: |
+          echo "cleaning up disk space..."
+          find . -type f -name '*.whl' -exec rm -rf {} \;
+          python -m pip cache purge
+          sudo rm -rf /usr/local/.ghcup
+          sudo rm -rf /opt/hostedtoolcache/CodeQL
+          sudo rm -rf /usr/local/lib/android/sdk/ndk
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf /usr/local/share/boost
+          if [[ "$(cat /etc/issue)" =~ Ubuntu ]]; then
+              sudo apt-get clean
+          fi
+          df -h
+      shell: bash
+
     - name: test
       id: test
       run: |

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/.github/workflows/test.yml

@@ -22,7 +22,9 @@ on:
       whl:
         description: "whl to test (variable appears late binding so unusable outside 'download artifact')"
         type: string
-        required: true
+      run_id:
+        description: run id of the BUILD job that generated the assets
+        type: string
 
   # makes workflow manually callable
   workflow_dispatch:
@@ -44,9 +46,11 @@ on:
         type: string
         required: true
       whl:
-        description: "whl to test (variable appears late binding so unusable outside 'download artifact')"
+        description: "whl to test (provide either whl or run_id)"
+        type: string
+      run_id:
+        description: run id of the BUILD job that generated the assets
         type: string
-        required: true
 
 jobs:
 
@@ -87,11 +91,33 @@ jobs:
 
             - name: download whl
               id: download
+              if: ${{ inputs.whl != '' }}
               uses: actions/download-artifact@v4
               with:
                   name: ${{ inputs.whl }}
                   path: ${{ inputs.whl }}
 
+            # GCP
+            - name: 'Authenticate to Google Cloud'
+              id: auth
+              uses: google-github-actions/auth@v2.1.3
+              with:
+                  project_id: ${{ secrets.GCP_PROJECT }}
+                  workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
+                  service_account: ${{ secrets.GCP_GHA_SA }}
+
+            - name: 'Set up Cloud SDK'
+              uses: 'google-github-actions/setup-gcloud@v2'
+              with:
+                  version: '>= 473.0.0'
+
+            - name: download assets
+              if: ${{ inputs.run_id != '' }}
+              uses: neuralmagic/nm-actions/actions/gcp-download-assets@v1.1.0
+              with:
+                  bucket_source: ${{ secrets.GCP_BUILD_ML_ASSETS2 }}
+                  run_id: ${{ inputs.run_id }}
+
             - name: run tests
               id: test
               uses: ./.github/actions/test/
@@ -109,20 +135,6 @@ jobs:
                   whl: ${{ inputs.whl }}
                   test_status: ${{ steps.test.outputs.status }}
 
-            # GCP
-            - name: 'Authenticate to Google Cloud'
-              id: auth
-              uses: google-github-actions/auth@v2.1.3
-              with:
-                  project_id: ${{ secrets.GCP_PROJECT }}
-                  workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}
-                  service_account: ${{ secrets.GCP_GHA_SA }}
-
-            - name: 'Set up Cloud SDK'
-              uses: 'google-github-actions/setup-gcloud@v2'
-              with:
-                  version: '>= 473.0.0'
-
             - name: copy results to GCP
               run: |
                   gcloud storage cp test-results/report.xml ${{ secrets.GCP_BUILD_ML_ASSETS2 }}/${{ github.run_id }}/test-results/report-${{ inputs.test_label }}.xml

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/.github/workflows/trigger-all.yml

@@ -32,8 +32,8 @@ jobs:
             wf_category: ${{ inputs.wf_category || 'NIGHTLY' }}
             gitref: ${{ inputs.gitref || 'main' }}
             push_to_pypi: ${{ (github.event.schedule == '30 0 * * *') || inputs.push_to_pypi || false }}
-            test_configs: '[{"python":"3.11.4","label":"ubuntu-22.04","timeout":"40"},
-                            {"python":"3.10.12","label":"ubuntu-24.04","timeout":"40"},
+            test_configs: '[{"python":"3.11.4","label":"ubuntu-24.04","timeout":"40"},
+                            {"python":"3.10.12","label":"ubuntu-22.04","timeout":"40"},
                             {"python":"3.9.17","label":"k8s-h100-solo","timeout":"40"},
                             {"python":"3.12.6","label":"k8s-a100-duo","timeout":"40"}]'
 

{compressed_tensors-0.10.3a20250620/src/compressed_tensors.egg-info → compressed_tensors-0.10.3a20250703}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250620
+Version: 0.10.3a20250703
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/src/compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.3.a20250620'
+__version__ = version = '0.10.3.a20250703'
 __version_tuple__ = version_tuple = (0, 10, 3)

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703/src/compressed_tensors.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250620
+Version: 0.10.3a20250703
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/src/compressed_tensors.egg-info/SOURCES.txt

@@ -130,6 +130,7 @@ tests/test_quantization/test_configs/__init__.py
 tests/test_quantization/test_configs/test_bit_depths.py
 tests/test_quantization/test_configs/test_strategies.py
 tests/test_quantization/test_utils/test_helpers.py
+tests/test_transform/conftest.py
 tests/test_transform/test_transform_args.py
 tests/test_transform/test_transform_config.py
 tests/test_transform/test_transform_scheme.py

compressed_tensors-0.10.3a20250703/tests/test_transform/conftest.py

@@ -0,0 +1,54 @@
+# Copyright (c) 2021 - present / Neuralmagic, Inc. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+import torch
+from compressed_tensors.transform import TransformArgs
+
+
+class TransformableModel(torch.nn.Module):
+    def __init__(self, *sizes):
+        super().__init__()
+        self.fcs = torch.nn.ModuleList(
+            [
+                torch.nn.Linear(sizes[index], sizes[index + 1], bias=False)
+                for index in range(0, len(sizes) - 1)
+            ]
+        )
+
+    def forward(self, x):
+        for layer in self.fcs:
+            x = layer(x)
+        return x
+
+
+@pytest.fixture(scope="function")
+def model_apply():
+    model = TransformableModel(2, 4, 8, 16, 32, 64)
+    apply = [
+        # weight output -> input
+        TransformArgs(targets="fcs.0", location="weight_output"),
+        TransformArgs(targets="fcs.1", location="input", inverse=True),
+        # output -> weight input
+        TransformArgs(targets="fcs.1", location="output"),
+        TransformArgs(targets="fcs.2", location="weight_input", inverse=True),
+        # output -> input
+        TransformArgs(targets="fcs.2", location="output"),
+        TransformArgs(targets="fcs.3", location="input", inverse=True),
+        # weight output -> weight input
+        TransformArgs(targets="fcs.3", location="weight_output"),
+        TransformArgs(targets="fcs.4", location="weight_input", inverse=True),
+    ]
+
+    return model, apply

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/tests/test_transform/factory/test_correctness.py

@@ -16,6 +16,7 @@ import pytest
 import torch
 from compressed_tensors.transform import (
     TransformArgs,
+    TransformConfig,
     TransformFactory,
     TransformScheme,
 )
@@ -23,27 +24,18 @@ from compressed_tensors.utils import offloaded_dispatch
 from tests.testing_utils import requires_accelerate, requires_gpu
 
 
-class TransformableModel(torch.nn.Module):
-    def __init__(self, *sizes):
-        super().__init__()
-        self.fcs = torch.nn.ModuleList([])
-        self.fcs.append(torch.nn.Linear(sizes[0], sizes[1], bias=False))
-        for index in range(1, len(sizes) - 1):
-            self.fcs.append(torch.nn.Linear(sizes[index], sizes[index + 1], bias=False))
+def scheme_kwargs():
+    all_types = TransformFactory.registered_names()
+    base = [{"type": type} for type in all_types]
+    randomized = [{"type": type, "randomize": True} for type in all_types]
+    return base + randomized
 
-    def forward(self, x):
-        for layer in self.fcs:
-            x = layer(x)
-        return x
 
-
-@pytest.mark.parametrize(
-    "scheme",
-    [TransformScheme(type=name) for name in TransformFactory.registered_names()],
-)
-def test_correctness_linear(scheme):
+@pytest.mark.parametrize("scheme_kwargs", scheme_kwargs())
+def test_correctness_linear(scheme_kwargs):
     size = (4, 8)
     module = torch.nn.Linear(*size, bias=True)
+    scheme = TransformScheme(**scheme_kwargs)
     factory = TransformFactory.from_scheme(scheme, name="")
 
     input_tfm = factory.create_transform(
@@ -67,50 +59,39 @@ def test_correctness_linear(scheme):
     assert torch.allclose(true_output, output, atol=1e-5, rtol=0.0)
 
 
-@pytest.mark.parametrize(
-    "scheme",
-    [TransformScheme(type=name) for name in TransformFactory.registered_names()],
-)
-def test_correctness_model(scheme, offload=False):
+@pytest.mark.parametrize("scheme_kwargs", scheme_kwargs())
+def test_correctness_model(scheme_kwargs, model_apply, offload=False):
     # load model
-    model = TransformableModel(2, 4, 8, 16, 32, 64)
+    model = model_apply[0]
     if offload:
         model = offloaded_dispatch(model, torch.device("cuda"))
 
-    # create factory
-    scheme.apply = [
-        # weight output -> input
-        TransformArgs(targets="fcs.0", location="weight_output"),
-        TransformArgs(targets="fcs.1", location="input", inverse=True),
-        # output -> weight input
-        TransformArgs(targets="fcs.1", location="output"),
-        TransformArgs(targets="fcs.2", location="weight_input", inverse=True),
-        # output -> input
-        TransformArgs(targets="fcs.2", location="output"),
-        TransformArgs(targets="fcs.3", location="input", inverse=True),
-        # weight output -> weight input
-        TransformArgs(targets="fcs.3", location="weight_output"),
-        TransformArgs(targets="fcs.4", location="weight_input", inverse=True),
-    ]
-    factory = TransformFactory.from_scheme(scheme, name="")
-
-    # create inputs
+    # get output
     input = torch.rand((17, model.fcs[0].in_features))
     if offload:
         input = input.to(torch.device("cuda"))
+    true_output = model(input)
+
+    # apply transforms
+    config = TransformConfig(
+        config_groups={
+            "": TransformScheme(
+                **scheme_kwargs,
+                apply=model_apply[1],
+            )
+        }
+    )
+    for name, scheme in config.config_groups.items():
+        factory = TransformFactory.from_scheme(scheme, name=name)
+        factory.apply_to_model(model)
 
     # compare outputs
-    true_output = model(input)
-    factory.apply_to_model(model)
     output = model(input)
     assert torch.allclose(true_output, output, atol=1e-5, rtol=0.0)
 
 
 @requires_gpu
 @requires_accelerate()
-@pytest.mark.parametrize(
-    "scheme",
-    [TransformScheme(type=name) for name in TransformFactory.registered_names()],
-)
-def test_correctness_model_offload(scheme):
-    test_correctness_model(scheme, offload=True)
+@pytest.mark.parametrize("scheme_kwargs", scheme_kwargs())
+def test_correctness_model_offload(scheme_kwargs, model_apply):
+    test_correctness_model(scheme_kwargs, model_apply, offload=True)

{compressed_tensors-0.10.3a20250620 → compressed_tensors-0.10.3a20250703}/tests/test_transform/factory/test_memory.py

@@ -19,49 +19,44 @@ import torch
 from compressed_tensors.transform import (
     TransformArgs,
     TransformBase,
+    TransformConfig,
     TransformFactory,
     TransformScheme,
 )
 from compressed_tensors.utils import align_modules, offloaded_dispatch
+from tests.test_transform.conftest import TransformableModel
 from tests.testing_utils import requires_accelerate, requires_gpu
 
 
-class TransformableModel(torch.nn.Module):
-    def __init__(self, *sizes):
-        super().__init__()
-        self.fcs = torch.nn.ModuleList([])
-        self.fcs.append(torch.nn.Linear(sizes[0], sizes[1], bias=False))
-        for index in range(1, len(sizes) - 1):
-            self.fcs.append(torch.nn.Linear(sizes[index], sizes[index + 1], bias=False))
+def scheme_kwargs():
+    all_types = TransformFactory.registered_names()
+    base = [{"type": type} for type in all_types]
+    randomized = [{"type": type, "randomize": True} for type in all_types]
+    return base + randomized
 
-    def forward(self, x):
-        for layer in self.fcs:
-            x = layer(x)
-        return x
-
-
-@pytest.mark.parametrize(
-    "scheme",
-    [TransformScheme(type=name) for name in TransformFactory.registered_names()],
-)
-def test_memory_sharing(scheme, offload=False):
-    # load scheme and factory
-    scheme = TransformScheme(
-        type="hadamard",
-        apply=[
-            TransformArgs(targets="Linear", location="input"),
-            TransformArgs(targets="Linear", location="output"),
-        ],
-    )
-    factory = TransformFactory.from_scheme(scheme, name="")
 
+@pytest.mark.parametrize("scheme_kwargs", scheme_kwargs())
+def test_memory_sharing(scheme_kwargs, offload=False):
     # load model (maybe with offloading)
     model = TransformableModel(2, 2, 4, 4, 8, 8)
     if offload:
         offloaded_dispatch(model, torch.device("cuda"))
 
     # add transforms to model
-    factory.apply_to_model(model)
+    config = TransformConfig(
+        config_groups={
+            "": TransformScheme(
+                **scheme_kwargs,
+                apply=[
+                    TransformArgs(targets="Linear", location="input"),
+                    TransformArgs(targets="Linear", location="output"),
+                ],
+            )
+        }
+    )
+    for name, scheme in config.config_groups.items():
+        factory = TransformFactory.from_scheme(scheme, name=name)
+        factory.apply_to_model(model)
 
     # check that memory is shared when onloaded
     with align_modules(model.modules()):
@@ -93,20 +88,12 @@ def test_memory_sharing(scheme, offload=False):
 
 @requires_gpu
 @requires_accelerate()
-@pytest.mark.parametrize(
-    "scheme",
-    [TransformScheme(type=name) for name in TransformFactory.registered_names()],
-)
-def test_memory_sharing_offload(scheme):
-    test_memory_sharing(scheme, offload=True)
+@pytest.mark.parametrize("scheme_kwargs", scheme_kwargs())
+def test_memory_sharing_offload(scheme_kwargs):
+    test_memory_sharing(scheme_kwargs, offload=True)
 
 
-@pytest.mark.parametrize(
-    "scheme",
-    [
-        TransformScheme(type=name, requires_grad=True)
-        for name in TransformFactory.registered_names()
-    ],
-)
-def test_memory_sharing_training(scheme):
-    test_memory_sharing(scheme, offload=False)
+@pytest.mark.parametrize("scheme_kwargs", scheme_kwargs())
+def test_memory_sharing_training(scheme_kwargs):
+    scheme_kwargs["requires_grad"] = True
+    test_memory_sharing(scheme_kwargs, offload=False)