optimum-rbln 0.8.0__tar.gz → 0.8.0.post2__tar.gz

{optimum_rbln-0.8.0 → optimum_rbln-0.8.0.post2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: optimum-rbln
-Version: 0.8.0
+Version: 0.8.0.post2
 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-0.8.0 → optimum_rbln-0.8.0.post2}/src/optimum/rbln/__version__.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.8.0'
-__version_tuple__ = version_tuple = (0, 8, 0)
+__version__ = version = '0.8.0.post2'
+__version_tuple__ = version_tuple = (0, 8, 0, 'post2')

{optimum_rbln-0.8.0 → optimum_rbln-0.8.0.post2}/src/optimum/rbln/configuration_utils.py

@@ -474,7 +474,15 @@ class RBLNModelConfig(RBLNSerializableConfigProtocol):
 
         if isinstance(submodule_config, dict):
             from_predecessor = self._runtime_options.copy()
+            from_predecessor.update(
+                {
+                    "npu": self.npu,
+                    "tensor_parallel_size": self.tensor_parallel_size,
+                    "optimum_rbln_version": self.optimum_rbln_version,
+                }
+            )
             from_predecessor.update(kwargs)
+
             init_kwargs = from_predecessor
             init_kwargs.update(submodule_config)
             submodule_config = submodule_config_cls(**init_kwargs)

{optimum_rbln-0.8.0 → optimum_rbln-0.8.0.post2}/src/optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py

@@ -306,7 +306,7 @@ class RBLNRuntimeModel(RBLNPytorchRuntime):
 
         # Pad input and cache_position if the last chunk is smaller than `prefill_chunk_size`
         if query_length % self.prefill_chunk_size != 0:
-            padding_size = self.prefill_chunk_size - query_length % self.prefill_chunk_size
+            padding_size = (self.prefill_chunk_size - query_length) % self.prefill_chunk_size
             # inputs_embeds
             if inputs.dim() == 3:
                 inputs = torch.nn.functional.pad(inputs, (0, 0, 0, padding_size))

{optimum_rbln-0.8.0 → optimum_rbln-0.8.0.post2}/src/optimum/rbln/transformers/models/gemma3/modeling_gemma3.py

@@ -479,8 +479,9 @@ class RBLNGemma3RuntimeModel(RBLNRuntimeModel):
             )
 
         # Pad input and cache_position if the last chunk is smaller than `prefill_chunk_size`
+        padding_size = 0
         if query_length % self.prefill_chunk_size != 0:
-            padding_size = self.prefill_chunk_size - query_length % self.prefill_chunk_size
+            padding_size = (self.prefill_chunk_size - query_length) % self.prefill_chunk_size
             # inputs_embeds
             if inputs.dim() == 3:
                 inputs = torch.nn.functional.pad(inputs, (0, 0, 0, padding_size))

{optimum_rbln-0.8.0 → optimum_rbln-0.8.0.post2}/src/optimum/rbln/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py

@@ -595,7 +595,7 @@ class RBLNQwen2_5_VLForConditionalGeneration(RBLNDecoderOnlyModelForCausalLM):
                 )
                 logits.append(output.logits)
             logits = torch.cat(logits, dim=0)
-        # Decoder
+            # Decoder
         else:
             inputs_embeds, position_embed = self._preprocess_decoder(input_ids, cache_position)
             output = self.decoder(

{optimum_rbln-0.8.0 → optimum_rbln-0.8.0.post2}/tests/test_config.py

@@ -2,8 +2,13 @@ import os
 import shutil
 
 import pytest
+import rebel
 
-from optimum.rbln import RBLNResNetForImageClassification, RBLNResNetForImageClassificationConfig
+from optimum.rbln import (
+    RBLNResNetForImageClassification,
+    RBLNResNetForImageClassificationConfig,
+    RBLNStableDiffusionPipeline,
+)
 from optimum.rbln.configuration_utils import RBLNCompileConfig
 
 
@@ -12,6 +17,39 @@ def model_id():
     return "hf-internal-testing/tiny-random-ResNetForImageClassification"
 
 
+@pytest.fixture
+def stable_diffusion_model():
+    model = RBLNStableDiffusionPipeline.from_pretrained(
+        "hf-internal-testing/tiny-sd-pipe",
+        export=True,
+        rbln_config={
+            "unet": {
+                "batch_size": 1,
+                "npu": "RBLN-CA22",
+                "create_runtimes": False,
+                "optimize_host_memory": False,
+            },
+            "text_encoder": {
+                "optimize_host_memory": False,
+            },
+        },
+    )
+    return model
+
+
+def test_stable_diffusion_config(stable_diffusion_model):
+    model = stable_diffusion_model
+    assert model is not None
+    assert model.unet.rbln_config.batch_size == 1
+    assert model.unet.rbln_config.npu == "RBLN-CA22"
+    assert model.unet.rbln_config.create_runtimes is False
+    assert model.unet.rbln_config.optimize_host_memory is False
+    assert model.unet.compiled_models[0]._meta["npu"] == "RBLN-CA22"
+
+    npu = rebel.get_npu_name()
+    assert model.text_encoder.compiled_models[0]._meta["npu"] == npu
+
+
 def test_explicit_config_parameters(model_id):
     """Test loading model with explicit configuration parameters."""
     model = RBLNResNetForImageClassification.from_pretrained(