optimum-rbln 0.9.2a5__tar.gz → 0.9.2a7__tar.gz

optimum_rbln-0.9.2a7/.github/version.yaml

@@ -0,0 +1 @@
+rebel_compiler_version: 0.9.2.dev164+g0896157e.prod

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/.github/workflows/check_code_quality.yml

@@ -12,7 +12,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: [3.9]
+        python-version: [3.12]
         os: [ubuntu-latest-rbln]
 
     runs-on: ${{ matrix.os }}

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/.github/workflows/rbln_check_compiler.yaml

@@ -25,7 +25,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.9"
+          python-version: "3.12"
 
       - name: Install rebel-compiler
         id: install_compiler

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/.github/workflows/rbln_optimum_inference_test.yaml

@@ -40,7 +40,7 @@ jobs:
           cache-dependency-glob: "uv.lock"
 
       - name: Setup Python
-        run: uv python install 3.9
+        run: uv python install 3.12
 
       - name: Install optimum-rbln with tests group dependencies
         run: |

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/.github/workflows/rbln_optimum_pytest.yaml

@@ -99,7 +99,7 @@ jobs:
 
       - name: Setup Python
         if: steps.should_skip.outputs.skip != 'true'
-        run: uv python install 3.9
+        run: uv python install 3.12
 
       - name: Install optimum-rbln with tests group dependencies
         if: steps.should_skip.outputs.skip != 'true'

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: optimum-rbln
-Version: 0.9.2a5
+Version: 0.9.2a7
 Summary: Optimum RBLN is the interface between the HuggingFace Transformers and Diffusers libraries and RBLN accelerators. It provides a set of tools enabling easy model loading and inference on single and multiple rbln device settings for different downstream tasks.
 Project-URL: Homepage, https://rebellions.ai
 Project-URL: Documentation, https://docs.rbln.ai

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/pyproject.toml

@@ -57,7 +57,7 @@ tests = [
     "soundfile>=0.13.1",
     "librosa>=0.11.0",
     "simphile>=1.0.2",
-    "torchcodec==0.5",
+    "torchcodec==0.7",
 ]
 quality = [
     "ruff>=0.3.3",

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/src/optimum/rbln/__version__.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.9.2a5'
-__version_tuple__ = version_tuple = (0, 9, 2, 'a5')
+__version__ = version = '0.9.2a7'
+__version_tuple__ = version_tuple = (0, 9, 2, 'a7')
 
 __commit_id__ = commit_id = None

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/src/optimum/rbln/modeling.py

@@ -249,37 +249,11 @@ class RBLNModel(RBLNBaseModel):
         trust_remote_code: bool = False,
         # Some rbln-config should be applied before loading torch module (i.e. quantized llm)
         rbln_config: Optional[RBLNModelConfig] = None,
-        dtype: Optional[Union[str, torch.dtype, dict]] = None,
         **kwargs,
     ) -> "PreTrainedModel":
         kwargs = cls.update_kwargs(kwargs)
 
-        hf_class = cls.get_hf_class()
-
-        if dtype is not None:
-            config = hf_class.config_class.from_pretrained(
-                model_id,
-                subfolder=subfolder,
-                revision=revision,
-                cache_dir=cache_dir,
-                use_auth_token=use_auth_token,
-                local_files_only=local_files_only,
-                force_download=force_download,
-                trust_remote_code=trust_remote_code,
-            )
-
-            config, processed_dtype, dtype_orig = _get_dtype(
-                cls=hf_class,
-                dtype=dtype,
-                config=config,
-            )
-
-            kwargs["torch_dtype"] = processed_dtype
-
-            if dtype_orig is not None:
-                hf_class._set_default_dtype(dtype_orig)
-
-        return hf_class.from_pretrained(
+        return cls.get_hf_class().from_pretrained(
             model_id,
             subfolder=subfolder,
             revision=revision,

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/src/optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py

@@ -322,22 +322,30 @@ class RBLNDecoderOnlyModel(RBLNModel, RBLNDecoderOnlyFlashAttentionMixin):
         *args,
         rbln_config: Optional[RBLNDecoderOnlyModelConfig] = None,
         num_hidden_layers: Optional[int] = None,
+        trust_remote_code: Optional[bool] = None,
+        torch_dtype: Optional[torch.dtype] = None,
+        dtype: Optional[torch.dtype] = None,
         **kwargs,
     ) -> PreTrainedModel:
         if rbln_config and rbln_config.quantization:
             model = cls.get_quantized_model(model_id, *args, rbln_config=rbln_config, **kwargs)
         else:
+            # TODO : resolve how to control PreTrainedConfig for hf_kwargs
             if num_hidden_layers is not None:
-                trust_remote_code = kwargs.get("trust_remote_code", None)
                 config, kwargs = AutoConfig.from_pretrained(
-                    model_id, return_unused_kwargs=True, num_hidden_layers=num_hidden_layers, **kwargs
+                    model_id,
+                    return_unused_kwargs=True,
+                    trust_remote_code=trust_remote_code,
+                    num_hidden_layers=num_hidden_layers,
+                    **kwargs,
                 )
                 if hasattr(config, "layer_types"):
                     config.layer_types = config.layer_types[:num_hidden_layers]
                 kwargs["config"] = config
-                kwargs["trust_remote_code"] = trust_remote_code
 
-            model = super().get_pytorch_model(model_id, *args, **kwargs)
+            model = super().get_pytorch_model(
+                model_id, *args, trust_remote_code=trust_remote_code, torch_dtype=torch_dtype, dtype=dtype, **kwargs
+            )
 
         return model
 

{optimum_rbln-0.9.2a5 → optimum_rbln-0.9.2a7}/uv.lock

@@ -1507,7 +1507,7 @@ tests = [
     { name = "sentencepiece", specifier = ">=0.2.0" },
     { name = "simphile", specifier = ">=1.0.2" },
     { name = "soundfile", specifier = ">=0.13.1" },
-    { name = "torchcodec", specifier = "==0.5" },
+    { name = "torchcodec", specifier = "==0.7" },
 ]
 
 [[package]]
@@ -2667,14 +2667,14 @@ wheels = [
 
 [[package]]
 name = "torchcodec"
-version = "0.5"
+version = "0.7.0"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/52/6e/d12c327338d9561f2ef0896218f163e6635d4ae1d688f1f28664d729b1ab/torchcodec-0.5-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:f760e49803d7767e10577d3d20d1a9dd0ad5adebd1ecbf7ce941176e8c388f54", size = 1366065, upload-time = "2025-07-23T17:42:41.88Z" },
-    { url = "https://files.pythonhosted.org/packages/d3/a1/cc2b7f72104cbb5f60f5f277f470e9d0172bffb354f74b0e13a02a241165/torchcodec-0.5-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:00252c2a138c655851cc7aa6e7ad6455c6513c9b2f93cec74563cba83196be8a", size = 1374385, upload-time = "2025-07-23T17:42:43.219Z" },
-    { url = "https://files.pythonhosted.org/packages/65/38/dfe8aa3a71c4eb0fd930d0c6db661cc65d1f220b2225433207ceb15733c4/torchcodec-0.5-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:310dfbfd7c56e283e0a9bd7cb4b5815cf1251a7bdeb93991b14311a1d144137d", size = 1370294, upload-time = "2025-07-23T17:42:44.455Z" },
-    { url = "https://files.pythonhosted.org/packages/71/e8/11be3eca1b545af1773277540eeafc5fd5a2b76433063756d6e61ea8bf8d/torchcodec-0.5-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:b9f41237de1b5960e85d3326a345deed86aa1b43d9cf17eac054255758dd41aa", size = 1375368, upload-time = "2025-07-23T17:42:45.619Z" },
-    { url = "https://files.pythonhosted.org/packages/af/db/94288001b785a0a388aedcdfeb7fce299b13537847ab67fe942b979bd73b/torchcodec-0.5-cp39-cp39-manylinux_2_28_x86_64.whl", hash = "sha256:47273264654747eef3519267e528c8a3855fbf7309b3b2038b42794d397675a8", size = 1366401, upload-time = "2025-07-23T17:42:47.353Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/81/cff42793544b7d3e2ff9a4912542c6d1c7a617aabe8404f8fd3d52453f20/torchcodec-0.7.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:a0071096724e8ded6a171457ce4680f646499b4a4d285cdb46e130983f965ce4", size = 1411823, upload-time = "2025-09-08T14:17:39.405Z" },
+    { url = "https://files.pythonhosted.org/packages/06/14/8ff28247988365fc47e8471e28cdfd8d037232fcf73abb67ee815ac80f1d/torchcodec-0.7.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:afb1c48b52bd4ee8f485f5a427bb4e82380590255a26b8e9e3fe099e0779287f", size = 1419444, upload-time = "2025-09-08T14:17:41.479Z" },
+    { url = "https://files.pythonhosted.org/packages/41/10/4a1a8407d0fad37cb43d1f749e7b422e5a0f6def17f3b90ab9ab9a105e32/torchcodec-0.7.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:3cd23c3296c9b071d56bb2c534a6a98275d65c1a6a7213cdb72a26ec9f9d2fd8", size = 1421871, upload-time = "2025-09-08T14:17:43.419Z" },
+    { url = "https://files.pythonhosted.org/packages/01/25/177ea01d138598ab68d5e3b000789e8617bf97874bd8f761d89093f419ba/torchcodec-0.7.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:9c70f910f9f48e6625aacaed534f766e13d447b895dc7299e96d4db9a93f1514", size = 1422493, upload-time = "2025-09-08T14:17:45.41Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/85/d4d3ac080325cdae3c8144de41afbf98bbf0f5e15e6c78644a70c5721ee6/torchcodec-0.7.0-cp39-cp39-manylinux_2_28_x86_64.whl", hash = "sha256:d2b10c1ad9bfdc21c75b1794566a0d68131f087dd609c27ee707efb035a45198", size = 1408250, upload-time = "2025-09-08T14:17:47.095Z" },
 ]
 
 [[package]]

optimum_rbln-0.9.2a5/.github/version.yaml

@@ -1 +0,0 @@
-rebel_compiler_version: 0.9.2.dev155+g1ab49983.prod