optimum-rbln 0.8.2a0__py3-none-any.whl → 0.9.3__py3-none-any.whl

optimum/rbln/ops/kv_cache_update.py

@@ -22,3 +22,8 @@ def rbln_cache_update(cache: Tensor, state: Tensor, position: Tensor, axis: Tens
     # This operation is designed to perform in-place updates directly on the device without needing to transfer the cache back to the host.
     # The `position` parameter specifies the start index for the update along the specified axis, allowing flexible updates to any part of the cache tensor.
     return torch.empty_like(cache)
+
+
+@rbln_cache_update.register_fake
+def rbln_cache_update_fake(cache: Tensor, state: Tensor, position: Tensor, axis: Tensor) -> Tensor:
+    return torch.empty_like(cache)

optimum/rbln/ops/linear.py

@@ -23,3 +23,10 @@ def linear(input: Tensor, weight: Tensor, bias: Optional[Tensor] = None) -> Tens
     output_shape = list(input.shape[:-1])
     output_shape += [weight.shape[0]]
     return torch.empty(size=output_shape, dtype=input.dtype, device=input.device, requires_grad=input.requires_grad)
+
+
+@linear.register_fake
+def linear_fake(input: Tensor, weight: Tensor, bias: Optional[Tensor] = None) -> Tensor:
+    output_shape = list(input.shape[:-1])
+    output_shape += [weight.shape[0]]
+    return torch.empty(size=output_shape, dtype=input.dtype, device=input.device, requires_grad=input.requires_grad)

optimum/rbln/transformers/__init__.py

@@ -34,6 +34,8 @@ _import_structure = {
         "RBLNAutoModelForSequenceClassification",
         "RBLNAutoModelForSpeechSeq2Seq",
         "RBLNAutoModelForVision2Seq",
+        "RBLNAutoModelForTextEncoding",
+        "RBLNAutoModelForZeroShotObjectDetection",
         "RBLNBartForConditionalGeneration",
         "RBLNBartForConditionalGenerationConfig",
         "RBLNBartModel",
@@ -52,6 +54,8 @@ _import_structure = {
         "RBLNBlip2VisionModelConfig",
         "RBLNColPaliForRetrieval",
         "RBLNColPaliForRetrievalConfig",
+        "RBLNColQwen2ForRetrieval",
+        "RBLNColQwen2ForRetrievalConfig",
         "RBLNCLIPTextModel",
         "RBLNCLIPTextModelConfig",
         "RBLNCLIPTextModelWithProjection",
@@ -62,12 +66,18 @@ _import_structure = {
         "RBLNCLIPVisionModelWithProjectionConfig",
         "RBLNDecoderOnlyModelForCausalLM",
         "RBLNDecoderOnlyModelForCausalLMConfig",
+        "RBLNDecoderOnlyModelConfig",
+        "RBLNDecoderOnlyModel",
         "RBLNDistilBertForQuestionAnswering",
         "RBLNDistilBertForQuestionAnsweringConfig",
         "RBLNDPTForDepthEstimation",
         "RBLNDPTForDepthEstimationConfig",
+        "RBLNDepthAnythingForDepthEstimation",
+        "RBLNDepthAnythingForDepthEstimationConfig",
         "RBLNExaoneForCausalLM",
         "RBLNExaoneForCausalLMConfig",
+        "RBLNGemmaModel",
+        "RBLNGemmaModelConfig",
         "RBLNGemma3ForCausalLM",
         "RBLNGemma3ForCausalLMConfig",
         "RBLNGemma3ForConditionalGeneration",
@@ -76,28 +86,66 @@ _import_structure = {
         "RBLNGemmaForCausalLMConfig",
         "RBLNGPT2LMHeadModel",
         "RBLNGPT2LMHeadModelConfig",
+        "RBLNGPT2Model",
+        "RBLNGPT2ModelConfig",
+        "RBLNGroundingDinoDecoder",
+        "RBLNGroundingDinoDecoderConfig",
+        "RBLNGroundingDinoForObjectDetection",
+        "RBLNGroundingDinoForObjectDetectionConfig",
+        "RBLNGroundingDinoEncoder",
+        "RBLNGroundingDinoEncoderConfig",
         "RBLNIdefics3ForConditionalGeneration",
         "RBLNIdefics3ForConditionalGenerationConfig",
         "RBLNIdefics3VisionTransformer",
         "RBLNIdefics3VisionTransformerConfig",
         "RBLNLlamaForCausalLM",
         "RBLNLlamaForCausalLMConfig",
+        "RBLNLlavaForConditionalGeneration",
+        "RBLNLlavaForConditionalGenerationConfig",
+        "RBLNLlamaModel",
+        "RBLNLlamaModelConfig",
+        "RBLNOPTForCausalLM",
+        "RBLNOPTForCausalLMConfig",
+        "RBLNPegasusForConditionalGeneration",
+        "RBLNPegasusForConditionalGenerationConfig",
+        "RBLNPegasusModel",
+        "RBLNPegasusModelConfig",
         "RBLNLlavaNextForConditionalGeneration",
         "RBLNLlavaNextForConditionalGenerationConfig",
+        "RBLNLoRAAdapterConfig",
+        "RBLNLoRAConfig",
         "RBLNMidmLMHeadModel",
         "RBLNMidmLMHeadModelConfig",
         "RBLNMistralForCausalLM",
         "RBLNMistralForCausalLMConfig",
+        "RBLNMistralModel",
+        "RBLNMistralModelConfig",
         "RBLNOPTForCausalLM",
         "RBLNOPTForCausalLMConfig",
+        "RBLNOPTModel",
+        "RBLNOPTModelConfig",
         "RBLNPhiForCausalLM",
         "RBLNPhiForCausalLMConfig",
+        "RBLNPixtralVisionModelConfig",
+        "RBLNPixtralVisionModel",
+        "RBLNPhiModel",
+        "RBLNPhiModelConfig",
         "RBLNQwen2_5_VisionTransformerPretrainedModel",
         "RBLNQwen2_5_VisionTransformerPretrainedModelConfig",
         "RBLNQwen2_5_VLForConditionalGeneration",
         "RBLNQwen2_5_VLForConditionalGenerationConfig",
+        "RBLNQwen2VisionTransformerPretrainedModel",
+        "RBLNQwen2VisionTransformerPretrainedModelConfig",
+        "RBLNQwen2VLForConditionalGeneration",
+        "RBLNQwen2VLForConditionalGenerationConfig",
+        "RBLNQwen2Model",
+        "RBLNQwen2ModelConfig",
         "RBLNQwen2ForCausalLM",
         "RBLNQwen2ForCausalLMConfig",
+        "RBLNQwen3ForCausalLM",
+        "RBLNQwen3ForCausalLMConfig",
+        "RBLNQwen3Model",
+        "RBLNQwen3ModelConfig",
         "RBLNResNetForImageClassification",
         "RBLNResNetForImageClassificationConfig",
         "RBLNRobertaForMaskedLM",
@@ -106,6 +154,8 @@ _import_structure = {
         "RBLNRobertaForSequenceClassificationConfig",
         "RBLNSiglipVisionModel",
         "RBLNSiglipVisionModelConfig",
+        "RBLNSwinBackbone",
+        "RBLNSwinBackboneConfig",
         "RBLNT5EncoderModel",
         "RBLNT5EncoderModelConfig",
         "RBLNT5ForConditionalGeneration",
@@ -141,7 +191,9 @@ if TYPE_CHECKING:
         RBLNAutoModelForSeq2SeqLM,
         RBLNAutoModelForSequenceClassification,
         RBLNAutoModelForSpeechSeq2Seq,
+        RBLNAutoModelForTextEncoding,
         RBLNAutoModelForVision2Seq,
+        RBLNAutoModelForZeroShotObjectDetection,
         RBLNBartForConditionalGeneration,
         RBLNBartForConditionalGenerationConfig,
         RBLNBartModel,
@@ -166,8 +218,16 @@ if TYPE_CHECKING:
         RBLNCLIPVisionModelConfig,
         RBLNCLIPVisionModelWithProjection,
         RBLNCLIPVisionModelWithProjectionConfig,
+        RBLNColPaliForRetrieval,
+        RBLNColPaliForRetrievalConfig,
+        RBLNColQwen2ForRetrieval,
+        RBLNColQwen2ForRetrievalConfig,
+        RBLNDecoderOnlyModel,
+        RBLNDecoderOnlyModelConfig,
         RBLNDecoderOnlyModelForCausalLM,
         RBLNDecoderOnlyModelForCausalLMConfig,
+        RBLNDepthAnythingForDepthEstimation,
+        RBLNDepthAnythingForDepthEstimationConfig,
         RBLNDistilBertForQuestionAnswering,
         RBLNDistilBertForQuestionAnsweringConfig,
         RBLNDPTForDepthEstimation,
@@ -180,30 +240,68 @@ if TYPE_CHECKING:
         RBLNGemma3ForConditionalGenerationConfig,
         RBLNGemmaForCausalLM,
         RBLNGemmaForCausalLMConfig,
+        RBLNGemmaModel,
+        RBLNGemmaModelConfig,
         RBLNGPT2LMHeadModel,
         RBLNGPT2LMHeadModelConfig,
+        RBLNGPT2Model,
+        RBLNGPT2ModelConfig,
+        RBLNGroundingDinoDecoder,
+        RBLNGroundingDinoDecoderConfig,
+        RBLNGroundingDinoEncoder,
+        RBLNGroundingDinoEncoderConfig,
+        RBLNGroundingDinoForObjectDetection,
+        RBLNGroundingDinoForObjectDetectionConfig,
         RBLNIdefics3ForConditionalGeneration,
         RBLNIdefics3ForConditionalGenerationConfig,
         RBLNIdefics3VisionTransformer,
         RBLNIdefics3VisionTransformerConfig,
         RBLNLlamaForCausalLM,
         RBLNLlamaForCausalLMConfig,
+        RBLNLlamaModel,
+        RBLNLlamaModelConfig,
+        RBLNLlavaForConditionalGeneration,
+        RBLNLlavaForConditionalGenerationConfig,
         RBLNLlavaNextForConditionalGeneration,
         RBLNLlavaNextForConditionalGenerationConfig,
+        RBLNLoRAAdapterConfig,
+        RBLNLoRAConfig,
         RBLNMidmLMHeadModel,
         RBLNMidmLMHeadModelConfig,
         RBLNMistralForCausalLM,
         RBLNMistralForCausalLMConfig,
+        RBLNMistralModel,
+        RBLNMistralModelConfig,
         RBLNOPTForCausalLM,
         RBLNOPTForCausalLMConfig,
+        RBLNOPTModel,
+        RBLNOPTModelConfig,
+        RBLNPegasusForConditionalGeneration,
+        RBLNPegasusForConditionalGenerationConfig,
+        RBLNPegasusModel,
+        RBLNPegasusModelConfig,
         RBLNPhiForCausalLM,
         RBLNPhiForCausalLMConfig,
+        RBLNPhiModel,
+        RBLNPhiModelConfig,
+        RBLNPixtralVisionModel,
+        RBLNPixtralVisionModelConfig,
         RBLNQwen2_5_VisionTransformerPretrainedModel,
         RBLNQwen2_5_VisionTransformerPretrainedModelConfig,
         RBLNQwen2_5_VLForConditionalGeneration,
         RBLNQwen2_5_VLForConditionalGenerationConfig,
         RBLNQwen2ForCausalLM,
         RBLNQwen2ForCausalLMConfig,
+        RBLNQwen2Model,
+        RBLNQwen2ModelConfig,
+        RBLNQwen2VisionTransformerPretrainedModel,
+        RBLNQwen2VisionTransformerPretrainedModelConfig,
+        RBLNQwen2VLForConditionalGeneration,
+        RBLNQwen2VLForConditionalGenerationConfig,
+        RBLNQwen3ForCausalLM,
+        RBLNQwen3ForCausalLMConfig,
+        RBLNQwen3Model,
+        RBLNQwen3ModelConfig,
         RBLNResNetForImageClassification,
         RBLNResNetForImageClassificationConfig,
         RBLNRobertaForMaskedLM,
@@ -212,6 +310,8 @@ if TYPE_CHECKING:
         RBLNRobertaForSequenceClassificationConfig,
         RBLNSiglipVisionModel,
         RBLNSiglipVisionModelConfig,
+        RBLNSwinBackbone,
+        RBLNSwinBackboneConfig,
         RBLNT5EncoderModel,
         RBLNT5EncoderModelConfig,
         RBLNT5ForConditionalGeneration,

optimum/rbln/transformers/configuration_generic.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Any, Dict, List, Optional, Tuple, Union
+from typing import Any, List, Optional, Tuple, Union
 
 from ..configuration_utils import RBLNModelConfig
 
@@ -25,7 +25,8 @@ class RBLNTransformerEncoderConfig(RBLNModelConfig):
         max_seq_len: Optional[int] = None,
         batch_size: Optional[int] = None,
         model_input_names: Optional[List[str]] = None,
-        **kwargs: Dict[str, Any],
+        model_input_shapes: Optional[List[Tuple[int, int]]] = None,
+        **kwargs: Any,
     ):
         """
         Args:
@@ -33,7 +34,7 @@ class RBLNTransformerEncoderConfig(RBLNModelConfig):
             batch_size (Optional[int]): The batch size for inference. Defaults to 1.
             model_input_names (Optional[List[str]]): Names of the input tensors for the model.
                 Defaults to class-specific rbln_model_input_names if not provided.
-            **kwargs: Additional arguments passed to the parent RBLNModelConfig.
+            kwargs: Additional arguments passed to the parent RBLNModelConfig.
 
         Raises:
             ValueError: If batch_size is not a positive integer.
@@ -45,6 +46,7 @@ class RBLNTransformerEncoderConfig(RBLNModelConfig):
             raise ValueError(f"batch_size must be a positive integer, got {self.batch_size}")
 
         self.model_input_names = model_input_names or self.rbln_model_input_names
+        self.model_input_shapes = model_input_shapes
 
 
 class RBLNImageModelConfig(RBLNModelConfig):
@@ -52,14 +54,14 @@ class RBLNImageModelConfig(RBLNModelConfig):
         self,
         image_size: Optional[Union[int, Tuple[int, int]]] = None,
         batch_size: Optional[int] = None,
-        **kwargs: Dict[str, Any],
+        **kwargs: Any,
     ):
         """
         Args:
             image_size (Optional[Union[int, Tuple[int, int]]]): The size of input images.
                 Can be an integer for square images or a tuple (height, width).
             batch_size (Optional[int]): The batch size for inference. Defaults to 1.
-            **kwargs: Additional arguments passed to the parent RBLNModelConfig.
+            kwargs: Additional arguments passed to the parent RBLNModelConfig.
 
         Raises:
             ValueError: If batch_size is not a positive integer.
@@ -116,30 +118,3 @@ class RBLNModelForImageClassificationConfig(RBLNImageModelConfig):
 
 class RBLNModelForDepthEstimationConfig(RBLNImageModelConfig):
     pass
-
-
-class RBLNModelForAudioClassificationConfig(RBLNModelConfig):
-    def __init__(
-        self,
-        batch_size: Optional[int] = None,
-        max_length: Optional[int] = None,
-        num_mel_bins: Optional[int] = None,
-        **kwargs: Dict[str, Any],
-    ):
-        """
-        Args:
-            batch_size (Optional[int]): The batch size for inference. Defaults to 1.
-            max_length (Optional[int]): Maximum length of the audio input in time dimension.
-            num_mel_bins (Optional[int]): Number of Mel frequency bins for audio processing.
-            **kwargs: Additional arguments passed to the parent RBLNModelConfig.
-
-        Raises:
-            ValueError: If batch_size is not a positive integer.
-        """
-        super().__init__(**kwargs)
-        self.batch_size = batch_size or 1
-        if not isinstance(self.batch_size, int) or self.batch_size < 0:
-            raise ValueError(f"batch_size must be a positive integer, got {self.batch_size}")
-
-        self.max_length = max_length
-        self.num_mel_bins = num_mel_bins