compressed-tensors 0.10.3a20250709__tar.gz → 0.10.3a20250711__tar.gz

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/.github/actions/test/action.yml

@@ -69,11 +69,11 @@ runs:
           echo "::endgroup::"
 
           if [[ "${ENABLE_COVERAGE}" == "true" ]]; then
-            echo "::group::consolidating coverage reports"
-            mkdir -p coverage-results
-            mv .coverage coverage-results/ || echo ".coverage file not found"
-            mv coverage-html coverage-results/ || echo "coverage-html folder not found"
-            mv coverage.json coverage-results/ || echo "coverage.json file not found"
+            echo "::group::check coverage reports"
+            if [ ! -d coverage-html ]; then
+                echo "ERROR: coverage-html folder not found"
+                exit 1
+            fi
             echo "::endgroup::"
           fi
           

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/.github/workflows/build-test.yml

@@ -25,7 +25,7 @@ on:
 
       # test related parameters
       test_configs:
-        description: "python, label, timeout"
+        description: "python, label, timeout, etc"
         type: string
         required: true
 
@@ -53,6 +53,7 @@ jobs:
             python: ${{ matrix.test_config.python }}
             timeout: ${{ matrix.test_config.timeout }}
             whl: ${{ needs.BUILD.outputs.whl }}
+            code_coverage: ${{ matrix.test_config.code_coverage || false }}
         secrets: inherit
 
     UPLOAD:

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/.github/workflows/test.yml

@@ -70,6 +70,10 @@ jobs:
         permissions:
             contents: 'read'
             id-token: 'write'
+            pages: 'write'
+        environment:
+            name: github-pages
+            url: ${{ steps.coverage.outputs.page_url }}
 
         steps:
 
@@ -134,6 +138,11 @@ jobs:
                   suitename: test-${{ inputs.python }}-${{ inputs.test_label }}
                   code_coverage: ${{ inputs.code_coverage }}
 
+            - name: extra info for summary
+              if: ${{ inputs.code_coverage }}
+              run: |
+                  echo "EXTRA='Code Coverage: https://neuralmagic.github.io/compressed-tensors/'" >> $GITHUB_ENV
+
             - name: summary
               uses: neuralmagic/nm-actions/actions/summary-test@v1.13.0
               if: success() || failure()
@@ -143,6 +152,7 @@ jobs:
                   python: ${{ inputs.python }}
                   whl: ${{ inputs.whl }}
                   test_status: ${{ steps.test.outputs.status }}
+                  extra: ${{ env.EXTRA }}
 
             - name: copy results to GCP
               run: |
@@ -157,9 +167,13 @@ jobs:
                   retention-days: 5
 
             - name: upload coverage report
-              uses: actions/upload-artifact@v4
-              if: (success() || failure()) && inputs.code_coverage
+              uses: actions/upload-pages-artifact@v3
+              if: ${{ inputs.code_coverage }}
               with:
-                  name: coverage-results
-                  path: coverage-results/*
+                  path: coverage-html
                   retention-days: 5
+
+            - name: deploy to Github Pages
+              id: coverage
+              uses: actions/deploy-pages@v4
+              if: ${{ inputs.code_coverage }}

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/.github/workflows/trigger-all.yml

@@ -32,7 +32,7 @@ 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-24.04","timeout":"40"},
+            test_configs: '[{"python":"3.11.4","label":"ubuntu-24.04","timeout":"40","code_coverage":true},
                             {"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-0.10.3a20250709/src/compressed_tensors.egg-info → compressed_tensors-0.10.3a20250711}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250709
+Version: 0.10.3a20250711
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/src/compressed_tensors/compressors/model_compressors/model_compressor.py

@@ -392,8 +392,8 @@ class ModelCompressor:
         for prefix, module in tqdm(model.named_modules(), desc="Compressing model"):
 
             if prefix in module_to_scheme or prefix in sparse_compression_targets:
-                module_device = get_execution_device(module).type
-                is_meta = (module_device == "meta")
+                module_device = get_execution_device(module)
+                is_meta = module_device.type == "meta"
 
                 exec_device = "meta" if is_meta else "cpu"
                 onloading_device = "meta" if is_meta else module_device
@@ -747,12 +747,16 @@ class ModelCompressor:
 
 def map_module_to_scheme(model: Module) -> Dict[str, QuantizationScheme]:
     """
-    Returns a dictionary which maps quantized module names to their quantization schemes
+    Returns a dictionary which maps quantized module names to their quantization
+    schemes. Only includes modules with weight quantization
     """
     return {
         fix_fsdp_module_name(name): module.quantization_scheme
         for name, module in model.named_modules()
-        if is_module_quantized(module)
+        if (
+            hasattr(module, "quantization_scheme") and
+            module.quantization_scheme.weights is not None
+        )
     }
 
 

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/src/compressed_tensors/compressors/sparse_compressors/sparse_24_bitmask.py

@@ -178,9 +178,13 @@ def sparse24_bitmask_compress(
 
     if tensor.is_meta:
         num_rows, num_cols = tensor.shape
-        compressed_values = torch.empty((num_rows, num_cols // 2), dtype=tensor.dtype, device="meta")
+        compressed_values = torch.empty(
+            (num_rows, num_cols // 2), dtype=tensor.dtype, device="meta"
+        )
         packed_cols = (num_cols + 7) // 8
-        bitmasks_packed = torch.empty((num_rows, packed_cols), dtype=torch.uint8, device="meta")
+        bitmasks_packed = torch.empty(
+            (num_rows, packed_cols), dtype=torch.uint8, device="meta"
+        )
         return compressed_values, bitmasks_packed
 
     bytemasks = get_24_bytemasks(tensor=tensor)

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/src/compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.3.a20250709'
+__version__ = version = '0.10.3.a20250711'
 __version_tuple__ = version_tuple = (0, 10, 3)

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711/src/compressed_tensors.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250709
+Version: 0.10.3a20250711
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250709 → compressed_tensors-0.10.3a20250711}/tests/test_compressors/model_compressors/test_model_compressor.py

@@ -446,10 +446,7 @@ def test_compress_model_meta(model_stub, q_format, s_config):
         cpu_model, s_config, q_format
     )
     # Only stores dtype because meta model does not store values
-    expected = {
-        k: v.dtype
-        for k, v in reference_compressor.compress(cpu_model).items()
-    }
+    expected = {k: v.dtype for k, v in reference_compressor.compress(cpu_model).items()}
 
     # Load model on meta device
     meta_model = AutoModelForCausalLM.from_pretrained(