optimum-rbln 0.9.1__py3-none-any.whl → 0.9.2a0__py3-none-any.whl

optimum/rbln/transformers/models/pixtral/modeling_pixtral.py

@@ -36,8 +36,6 @@ logger = get_logger(__name__)
 if TYPE_CHECKING:
     from transformers import AutoFeatureExtractor, AutoProcessor, AutoTokenizer, PreTrainedModel
 
-    from ....diffusers.modeling_diffusers import RBLNDiffusionMixin, RBLNDiffusionMixinConfig
-
 
 class RBLNRuntimePixtralVisionModel(RBLNPytorchRuntime):
     mandatory_members = ["main_input_name"]
@@ -128,8 +126,11 @@ class RBLNRuntimePixtralVisionModel(RBLNPytorchRuntime):
                 (1, patch_embed_seq.shape[-2]), fill_value=torch.finfo(patch_embed_seq.dtype).min
             )
             attention_mask[:, : h_patched_original * w_patched_original] = 0
-
-            transformer_output = super().forward(patch_embed_seq, attention_mask, cos, sin)
+            if "out" in kwargs:
+                super().forward(patch_embed_seq, attention_mask, cos, sin, **kwargs)
+                transformer_output = kwargs["out"]
+            else:
+                transformer_output = super().forward(patch_embed_seq, attention_mask, cos, sin, **kwargs)
 
             last_hidden_state_list.append(transformer_output[0][:, : h_patched_original * w_patched_original, :])
             hidden_states = transformer_output[1:]
@@ -236,12 +237,6 @@ class RBLNPixtralVisionModel(RBLNModel):
         }
         return _PixtralVisionModel(model, **wrapper_cfg).eval()
 
-    @classmethod
-    def update_rbln_config_using_pipe(
-        cls, pipe: "RBLNDiffusionMixin", rbln_config: "RBLNDiffusionMixinConfig", submodule_name: str
-    ) -> "RBLNDiffusionMixinConfig":
-        return rbln_config
-
     @classmethod
     def _update_rbln_config(
         cls,
@@ -309,7 +304,7 @@ class RBLNPixtralVisionModel(RBLNModel):
             )
 
         output = self.model(
-            pixel_values, image_sizes, output_hidden_states=output_hidden_states, return_dict=return_dict
+            pixel_values, image_sizes, output_hidden_states=output_hidden_states, return_dict=return_dict, **kwargs
         )
 
         return output

optimum/rbln/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py

@@ -23,6 +23,7 @@ from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import (
     Qwen2_5_VisionPatchEmbed,
     Qwen2_5_VisionRotaryEmbedding,
     Qwen2_5_VisionTransformerPretrainedModel,
+    Qwen2_5_VLModel,
     Qwen2_5_VLRotaryEmbedding,
 )
 
@@ -391,6 +392,14 @@ class RBLNQwen2_5_VLForConditionalGeneration(RBLNDecoderOnlyModelForCausalLM):
     def can_generate(self):
         return True
 
+    @classmethod
+    def get_pytorch_model(cls, *args, **kwargs):
+        model = super().get_pytorch_model(*args, **kwargs)
+        model.model.lm_head = model.lm_head
+        model.lm_head = None
+        del model.lm_head
+        return model
+
     @classmethod
     def update_kwargs(cls, kwargs):
         kwargs.update(
@@ -532,7 +541,8 @@ class RBLNQwen2_5_VLForConditionalGeneration(RBLNDecoderOnlyModelForCausalLM):
             vision_tokens = input_id[0][vision_start_indices + 1]
             image_nums = (vision_tokens == image_token_id).sum()
             video_nums = (vision_tokens == video_token_id).sum()
-            position_ids, rope_deltas = self.get_rope_index(
+            position_ids, rope_deltas = Qwen2_5_VLModel.get_rope_index(
+                self,
                 input_id,
                 image_grid_thw[image_idx : image_idx + image_nums] if image_grid_thw is not None else None,
                 video_grid_thw[video_idx : video_idx + video_nums] if video_grid_thw is not None else None,

optimum/rbln/transformers/models/qwen2_5_vl/qwen2_5_vl_architecture.py

@@ -3,6 +3,7 @@ from typing import Tuple
 
 import torch
 import torch.nn as nn
+from transformers import PreTrainedModel
 
 from ..decoderonly.decoderonly_architecture import DecoderOnlyWrapper, apply_rotary_pos_emb
 
@@ -194,3 +195,24 @@ class Qwen2_5_VL_LanguageModelWrapper(DecoderOnlyWrapper):
             past_key_values,
             position_embeds,
         )
+
+    def convert_to_rbln_class(self, model: PreTrainedModel, max_seq_len: int):
+        new_layers = []
+
+        for layer_idx, layer in enumerate(model.model.language_model.layers):
+            is_sliding = layer_idx in self.rbln_config.sliding_window_layers
+            new_self_attn = self.get_rbln_attn_class()(
+                self.get_attn_layer(layer), self.rbln_config, is_sliding=is_sliding
+            )
+            new_layer = self.get_rbln_layer_class()(layer, new_self_attn)
+            new_layers.append(new_layer)
+
+        new_model = self.get_rbln_model_class()(
+            model.model.language_model,
+            new_layers,
+            self.rbln_config,
+            use_learned_pos_emb=self.__class__._use_learned_pos_emb,
+        )
+
+        new_model = self.get_rbln_causal_lm_class()(model.model, new_model)
+        return new_model

optimum/rbln/transformers/models/qwen2_vl/modeling_qwen2_vl.py

@@ -27,6 +27,7 @@ from transformers.modeling_utils import no_init_weights
 from transformers.models.qwen2_vl.modeling_qwen2_vl import (
     PatchEmbed,
     Qwen2VisionTransformerPretrainedModel,
+    Qwen2VLModel,
     Qwen2VLRotaryEmbedding,
     VisionRotaryEmbedding,
 )
@@ -280,6 +281,14 @@ class RBLNQwen2VLForConditionalGeneration(RBLNDecoderOnlyModelForCausalLM):
     def can_generate(self):
         return True
 
+    @classmethod
+    def get_pytorch_model(cls, *args, **kwargs):
+        model = super().get_pytorch_model(*args, **kwargs)
+        model.model.lm_head = model.lm_head
+        model.lm_head = None
+        del model.lm_head
+        return model
+
     @classmethod
     def get_input_info(
         cls,
@@ -402,7 +411,8 @@ class RBLNQwen2VLForConditionalGeneration(RBLNDecoderOnlyModelForCausalLM):
             vision_tokens = input_id[0][vision_start_indices + 1]
             image_nums = (vision_tokens == image_token_id).sum()
             video_nums = (vision_tokens == video_token_id).sum()
-            position_ids, rope_deltas = self.get_rope_index(
+            position_ids, rope_deltas = Qwen2VLModel.get_rope_index(
+                self,
                 input_id,
                 image_grid_thw[image_idx : image_idx + image_nums] if image_grid_thw is not None else None,
                 video_grid_thw[video_idx : video_idx + video_nums] if video_grid_thw is not None else None,

optimum/rbln/transformers/models/qwen2_vl/qwen2_vl_architecture.py

@@ -3,6 +3,7 @@ from typing import Tuple
 
 import torch
 import torch.nn as nn
+from transformers import PreTrainedModel
 
 from ..decoderonly.decoderonly_architecture import (
     DecoderOnlyWrapper,
@@ -139,3 +140,24 @@ class Qwen2VL_LanguageModelWrapper(DecoderOnlyWrapper):
             past_key_values,
             position_embeds,
         )
+
+    def convert_to_rbln_class(self, model: PreTrainedModel, max_seq_len: int):
+        new_layers = []
+
+        for layer_idx, layer in enumerate(model.model.language_model.layers):
+            is_sliding = layer_idx in self.rbln_config.sliding_window_layers
+            new_self_attn = self.get_rbln_attn_class()(
+                self.get_attn_layer(layer), self.rbln_config, is_sliding=is_sliding
+            )
+            new_layer = self.get_rbln_layer_class()(layer, new_self_attn)
+            new_layers.append(new_layer)
+
+        new_model = self.get_rbln_model_class()(
+            model.model.language_model,
+            new_layers,
+            self.rbln_config,
+            use_learned_pos_emb=self.__class__._use_learned_pos_emb,
+        )
+
+        new_model = self.get_rbln_causal_lm_class()(model.model, new_model)
+        return new_model

optimum/rbln/transformers/models/siglip/modeling_siglip.py

@@ -29,8 +29,6 @@ logger = get_logger(__name__)
 if TYPE_CHECKING:
     from transformers import AutoFeatureExtractor, AutoProcessor, AutoTokenizer, PreTrainedModel
 
-    from ....diffusers.modeling_diffusers import RBLNDiffusionMixin, RBLNDiffusionMixinConfig
-
 
 class _SiglipVisionModel(torch.nn.Module):
     def __init__(
@@ -65,6 +63,8 @@ class RBLNSiglipVisionModel(RBLNModel):
     on RBLN devices, supporting image encoding for multimodal vision-language tasks.
     """
 
+    _tp_support = False
+
     @classmethod
     def wrap_model_if_needed(cls, model: torch.nn.Module, rbln_config: RBLNSiglipVisionModelConfig) -> torch.nn.Module:
         wrapper_cfg = {
@@ -74,12 +74,6 @@ class RBLNSiglipVisionModel(RBLNModel):
         }
         return _SiglipVisionModel(model, **wrapper_cfg).eval()
 
-    @classmethod
-    def update_rbln_config_using_pipe(
-        cls, pipe: "RBLNDiffusionMixin", rbln_config: "RBLNDiffusionMixinConfig", submodule_name: str
-    ) -> "RBLNDiffusionMixinConfig":
-        return rbln_config
-
     @classmethod
     def _update_rbln_config(
         cls,
@@ -128,11 +122,6 @@ class RBLNSiglipVisionModel(RBLNModel):
         interpolate_pos_encoding: bool = False,
         **kwargs: Any,
     ) -> Union[Tuple, BaseModelOutputWithPooling]:
-        if len(kwargs) > 0 and any(value is not None for value in kwargs.values()):
-            logger.warning(
-                f"Currently, optimum-rbln does not support kwargs {kwargs.keys()} for {self.__class__.__name__}."
-            )
-
         output_attentions = output_attentions if output_attentions is not None else self.rbln_config.output_attentions
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.rbln_config.output_hidden_states
@@ -156,7 +145,7 @@ class RBLNSiglipVisionModel(RBLNModel):
                 f"Please compile again with the correct argument."
             )
 
-        output = super().forward(pixel_values, return_dict=return_dict)
+        output = super().forward(pixel_values, return_dict=return_dict, **kwargs)
         return output
 
     def _prepare_output(self, output, return_dict):

optimum/rbln/transformers/utils/rbln_runtime_wrapper.py

@@ -0,0 +1,79 @@
+# Copyright 2025 Rebellions 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.
+
+
+from abc import ABC, abstractmethod
+from typing import TYPE_CHECKING, Any, Dict, List, Tuple, Union
+
+from torch.nn import Module
+
+from ...modeling import RBLNModel
+
+
+if TYPE_CHECKING:
+    import rebel
+
+
+class LoopProcessor(Module, ABC):
+    def __init__(self, model: Union[RBLNModel, "rebel.Runtime"]):
+        super().__init__()
+        self.model = model
+
+    def __repr__(self) -> str:
+        return repr(self.model)
+
+    def _is_batch_implemented(self) -> bool:
+        return self._forward_batch.__func__ is not LoopProcessor._forward_batch
+
+    def forward(self, *args, force_loop: bool = False, **kwargs) -> Any:
+        if not force_loop and self._is_batch_implemented():
+            return self._forward_batch(*args, **kwargs)
+        else:
+            return self._forward_loop(*args, **kwargs)
+
+    def _forward_loop(self, *args, **kwargs) -> Any:
+        batch_size = self._get_batch_size(*args, **kwargs)
+
+        if not isinstance(batch_size, int) or batch_size == 0:
+            return self._process_outputs([])
+
+        common_inputs = self._prepare_inputs_before_loop(*args, **kwargs)
+
+        outputs = []
+        for i in range(batch_size):
+            item_args, item_kwargs = self._prepare_inputs_for_iteration(i, common_inputs, *args, **kwargs)
+            item_output = self.model(*item_args, **item_kwargs)
+            outputs.append(item_output)
+
+        return self._process_outputs(outputs, **kwargs)
+
+    def _forward_batch(self, *args, **kwargs) -> Any:
+        raise NotImplementedError("The batch processing logic (_forward_batch) is not implemented in this class.")
+
+    @abstractmethod
+    def _get_batch_size(self, *args, **kwargs) -> int:
+        pass
+
+    @abstractmethod
+    def _prepare_inputs_for_iteration(
+        self, index: int, common_inputs: Dict[str, Any], *args, **kwargs
+    ) -> Tuple[List[Any], Dict[str, Any]]:
+        pass
+
+    def _prepare_inputs_before_loop(self, *args, **kwargs) -> Dict[str, Any]:
+        pass
+
+    @abstractmethod
+    def _process_outputs(self, outputs: List[Any], **kwargs) -> Any:
+        pass

optimum/rbln/utils/submodule.py

@@ -17,7 +17,7 @@ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Type, Union
 
 from transformers import PretrainedConfig
 
-from ..configuration_utils import RBLNModelConfig
+from ..configuration_utils import RBLNModelConfig, get_rbln_config_class
 from ..utils.model_utils import get_rbln_model_cls
 
 
@@ -41,6 +41,15 @@ class SubModulesMixin:
         for submodule_meta, submodule in zip(self._rbln_submodules, rbln_submodules):
             setattr(self, submodule_meta["name"], submodule)
 
+    @classmethod
+    def _get_submodule_config_class(
+        cls, cls_name: str, submodule_rbln_config: Dict[str, Any]
+    ) -> Type[RBLNModelConfig]:
+        if isinstance(submodule_rbln_config, dict) and "cls_name" in submodule_rbln_config:
+            config_cls_name = submodule_rbln_config["cls_name"]
+            return get_rbln_config_class(config_cls_name)
+        return get_rbln_config_class(f"RBLN{cls_name}Config")
+
     @classmethod
     def _update_submodule_config(
         cls,
@@ -69,12 +78,19 @@ class SubModulesMixin:
             cls_name = torch_submodule.__class__.__name__
             submodule_cls: Type["RBLNModel"] = get_rbln_model_cls(f"RBLN{cls_name}")
             submodule_rbln_config = getattr(rbln_config, submodule_name) or {}
+            submodule_config_cls = cls._get_submodule_config_class(cls_name, submodule_rbln_config)
 
             if isinstance(submodule_rbln_config, dict):
-                submodule_rbln_config_class = submodule_cls.get_rbln_config_class()
-                submodule_rbln_config = submodule_rbln_config_class(**submodule_rbln_config)
-                setattr(rbln_config, submodule_name, submodule_rbln_config)
-
+                filtered_kwargs = rbln_config.filter_parameters(submodule_config_cls, submodule_rbln_config)
+                filtered_kwargs["cls_name"] = submodule_config_cls.__name__
+                submodule_rbln_config = submodule_config_cls(**filtered_kwargs)
+            elif not isinstance(submodule_rbln_config, submodule_config_cls):
+                config_dict = {k: v for k, v in submodule_rbln_config.__dict__.items() if not k.startswith("_")}
+                filtered_kwargs = rbln_config.filter_parameters(submodule_config_cls, config_dict)
+                filtered_kwargs["cls_name"] = submodule_config_cls.__name__
+                submodule_rbln_config = submodule_config_cls(**filtered_kwargs)
+
+            setattr(rbln_config, submodule_name, submodule_rbln_config)
             submodule_rbln_config = submodule_cls._update_submodule_config(model, submodule_rbln_config, preprocessors)
 
             rbln_submodule = submodule_cls.from_model(

{optimum_rbln-0.9.1.dist-info → optimum_rbln-0.9.2a0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: optimum-rbln
-Version: 0.9.1
+Version: 0.9.2a0
 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
@@ -29,7 +29,7 @@ Requires-Dist: packaging>=24.1
 Requires-Dist: torch==2.7.0
 Requires-Dist: torchaudio<=2.7.0
 Requires-Dist: torchvision<=0.22.0
-Requires-Dist: transformers==4.51.3
+Requires-Dist: transformers==4.53.1
 Description-Content-Type: text/markdown
 
 

{optimum_rbln-0.9.1.dist-info → optimum_rbln-0.9.2a0.dist-info}/RECORD

@@ -1,6 +1,6 @@
 optimum/rbln/__init__.py,sha256=DAJM5PWAYFiWVlyxVXUvj3CaFOEhX1yhEfhIt1LxL-A,18714
-optimum/rbln/__version__.py,sha256=LwGndsRSpclYq-j3wgRr2nzOXwUYj0Jtg7Kof7R0BEw,704
-optimum/rbln/configuration_utils.py,sha256=XYXqbriu7DZkoCqicM1iXzqrSd7BVO-e2vA-8WSGgN0,36248
+optimum/rbln/__version__.py,sha256=N_fc72G3EaGWKHzHnNEtPoHLzurnsMp0Or2eVdiYVTM,712
+optimum/rbln/configuration_utils.py,sha256=nAzDwa_2-X32bikEJ2zbRZIk68s50hAP7gTmKN07Ig0,38080
 optimum/rbln/modeling.py,sha256=IZ8loagxm--2BcqTl16KRHUR3hkccpeaY2grOWOtwqk,14473
 optimum/rbln/modeling_base.py,sha256=poXfHZCAlFd28MY9dvMi7tC2RytLx77Lee2XGS_KeZg,27684
 optimum/rbln/diffusers/__init__.py,sha256=1tgU_xWA42BmInqu9bBz_5R_E9TGhhK3mI06YlaiTLg,7232
@@ -16,12 +16,12 @@ optimum/rbln/diffusers/configurations/models/configuration_transformer_sd3.py,sh
 optimum/rbln/diffusers/configurations/models/configuration_unet_2d_condition.py,sha256=11kOnB-o5MjCJHDQvEhGxXpBNZLC1r3HU3-wiZ_3bL0,3611
 optimum/rbln/diffusers/configurations/models/configuration_vq_model.py,sha256=GbJKrTOUdvpcGQ786IIKF0rqkRrvZh_E5n3o9bLN8uY,3240
 optimum/rbln/diffusers/configurations/pipelines/__init__.py,sha256=RfJXQiYvgGc3Rp7JYk5s0AQd0XB5JCAb37_riGWQAYg,1268
-optimum/rbln/diffusers/configurations/pipelines/configuration_controlnet.py,sha256=6ObO1b3ZAwf9e6oQOeHh0sk8NzFXayDCXmxDBx-uNwc,14183
-optimum/rbln/diffusers/configurations/pipelines/configuration_cosmos.py,sha256=eirI9X0YubVQ85hd-BEH77I8SZ9hlnPJGGNpj-MO3eQ,4596
-optimum/rbln/diffusers/configurations/pipelines/configuration_kandinsky2_2.py,sha256=3vDGOgd3sNhh3D5mXVAljXXiwiUyZz7h9Z5752uB880,16341
-optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion.py,sha256=6kvLFH1vN59Pc_rHAyEcT3KXoqulxFiaFcZakyBv9hs,6572
-optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_3.py,sha256=qkbSG6eNhDQK7z-4FOBKf-zSsEdjtQH9PHkvlAuk1zE,7904
-optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_xl.py,sha256=hZOr0nVApS-HrwVim6S2_5gO0Zb8fburnaJXe8Xvqe4,7020
+optimum/rbln/diffusers/configurations/pipelines/configuration_controlnet.py,sha256=ALwMwJret_KQKXld-Y-ZRJh_Wp4qza1xZsbOEUJutg0,14525
+optimum/rbln/diffusers/configurations/pipelines/configuration_cosmos.py,sha256=niblhj4brobRCHF0NoVfdIhChUsup2O9IvFddnopIUc,4701
+optimum/rbln/diffusers/configurations/pipelines/configuration_kandinsky2_2.py,sha256=_lNqy0OFA4sIsn3tOvYwR6X_-y3JTN67ePTWBkt2pFc,16598
+optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion.py,sha256=NnisLdUkW3QQfzn8ZtWUbBwXZF6JF3vj9UI0MthH0EM,6625
+optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_3.py,sha256=vDb7vqH2XycB86L5JH-ofXfdngU2BseJvVw7YXrzmgg,8039
+optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_xl.py,sha256=qX6-HvOt8SBstEeigWWcor-z2bmyoqAucjRirNfma5o,7161
 optimum/rbln/diffusers/models/__init__.py,sha256=4xHGuJ0HmZXBt6Xc8rVmGKvvQ366yxJarV5Vg_4-cso,1541
 optimum/rbln/diffusers/models/controlnet.py,sha256=6owledPe9BXhbZOG8lbuuYvpBU0UrQV7zmat6SoMXOM,10585
 optimum/rbln/diffusers/models/autoencoders/__init__.py,sha256=kpINW4bWwy-Q2doPME4nZ8gXRmkK2DRv2kDdbZuQ3m8,738
@@ -44,7 +44,7 @@ optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py,sha25
 optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py,sha256=2w6dmGQuBWqVoocn27z2yMkG7fL7_MVDBcQNJPJsRXU,45300
 optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py,sha256=HX56itORMqXLjZcwv25C-_z3JyZn3v6BpfIjsrDO3mE,46640
 optimum/rbln/diffusers/pipelines/cosmos/__init__.py,sha256=h2j6S8IJPVHeNU8qmW9vyXMgHBw0d7kQcuMAA5YoHPU,795
-optimum/rbln/diffusers/pipelines/cosmos/configuration_cosmos_guardrail.py,sha256=NsEaRYgiddopL9OE_9CsZqzbikkzoxUoRQaRML-Jqrw,3709
+optimum/rbln/diffusers/pipelines/cosmos/configuration_cosmos_guardrail.py,sha256=BRHWXCxRO5l0YTFTBSkS4sd5Hx8YeuHB14Fr2copYFY,3747
 optimum/rbln/diffusers/pipelines/cosmos/cosmos_guardrail.py,sha256=Hmklj-sGNtUst6sTLLDXyQzt7zFoNIHytZpPQ-V1CPw,16676
 optimum/rbln/diffusers/pipelines/cosmos/pipeline_cosmos_text2world.py,sha256=biveGIpJVGL0rH_YLgynvt47QQA41QdjRtY7G32jDXY,5638
 optimum/rbln/diffusers/pipelines/cosmos/pipeline_cosmos_video2world.py,sha256=96dHWMIe1x_Q3G64fsNhulfjmd_PrSXRt0GwgNjYMlg,5645
@@ -94,21 +94,21 @@ optimum/rbln/transformers/models/bert/bert_architecture.py,sha256=cZgf-B-FV8qbeJ
 optimum/rbln/transformers/models/bert/configuration_bert.py,sha256=nEZnX6LXpLKWaoPEd4pWSysw9h-PLb2ld0ibC3dcJ7w,1611
 optimum/rbln/transformers/models/bert/modeling_bert.py,sha256=7MQZS11k4__oyeni5ek2SzRf-gtD3_hMKl_oOzN7_XQ,2263
 optimum/rbln/transformers/models/blip_2/__init__.py,sha256=L01gPXcUCa8Vg-bcng20vZvBIN_jlqCzwUSFuq0QOag,855
-optimum/rbln/transformers/models/blip_2/configuration_blip_2.py,sha256=2pr8yPaV_GeQuOlTLceS7Ph18x4Kd7D6hbsLu75Hta8,3355
-optimum/rbln/transformers/models/blip_2/modeling_blip_2.py,sha256=eJEKk0TRHT9kqoG80xFlGo3PGamzoBDoQSZJPXmvgPE,16283
+optimum/rbln/transformers/models/blip_2/configuration_blip_2.py,sha256=8eSilBwcPWQhBg-oilCmDPo-DN6V5lpLMlTB7WPknII,4630
+optimum/rbln/transformers/models/blip_2/modeling_blip_2.py,sha256=_IGDGS7VJfw5Qi5AJYCjduRVlWHGADhSAiTY42Jppys,16389
 optimum/rbln/transformers/models/clip/__init__.py,sha256=TLeXDqcFK6M6v9x7Xr64kBbqGu3hFHM7p754dQ8UVQc,938
 optimum/rbln/transformers/models/clip/configuration_clip.py,sha256=Ea8TCVmMayydfw9p4kTP3UdtvoaPWf4Z4claB61JuE4,4175
-optimum/rbln/transformers/models/clip/modeling_clip.py,sha256=sIjZz1HtINZFDFyimbUVgbdUzxmoXHG3mqA5ROTir9w,13148
+optimum/rbln/transformers/models/clip/modeling_clip.py,sha256=BLAYJAtv_2ZnKOlZ8iDBr2Su3bKM_eMWeUSK9MOaj7I,13198
 optimum/rbln/transformers/models/colpali/__init__.py,sha256=n3rueXT_oC0N8myoZiic0YkVK24CW5hZBPa-0L8so6Y,119
-optimum/rbln/transformers/models/colpali/colpali_architecture.py,sha256=Tl3_fA7faM3odc-bYp-zGjvBSsDFcrJ1m1d8ogFnk7s,8052
-optimum/rbln/transformers/models/colpali/configuration_colpali.py,sha256=U_UeaBnD_sQoDQznnbO1_SCHwz3yt3bnVJF1My0krxQ,2617
-optimum/rbln/transformers/models/colpali/modeling_colpali.py,sha256=Qw6REaRFFj0_BYGsUn-rLQY7eEi8QAioYwEqLKUdTXE,17542
+optimum/rbln/transformers/models/colpali/colpali_architecture.py,sha256=TCOW3v5l9fIt1uIFtWa8ZAxq1cdCER8gXWjmbLQD20M,8079
+optimum/rbln/transformers/models/colpali/configuration_colpali.py,sha256=_HuZBVV-ponml95UapkYpRhffZy53-9jSZknx7hID7o,3348
+optimum/rbln/transformers/models/colpali/modeling_colpali.py,sha256=YnAUMTbcX2idRLMkZ3Qiv-GZFodfCun5-3sDfiIz77s,17955
 optimum/rbln/transformers/models/decoderonly/__init__.py,sha256=w3VZOIBYaHXVdnuhK4y0zWAj0IAv7_5LGTJYaz9oYmI,1056
 optimum/rbln/transformers/models/decoderonly/configuration_decoderonly.py,sha256=yUhB9yMFVbq7B9WtAM6m8cQ4K0U3a2N-TU34N5zvjRM,15180
 optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py,sha256=ySsiU0Acj5enJW-SqtFMAfBeH0HeqlhCd78QlpKJNQw,42780
 optimum/rbln/transformers/models/decoderonly/decoderonly_runtime_utils.py,sha256=v3mfIlQImQkYYr-rPn7rQR3GYdVUhALRttEduLI7H9c,20012
 optimum/rbln/transformers/models/decoderonly/generation_decoderonly.py,sha256=zabSgQd2VzHhkpbhUFW5Z-CjYB1JvSJOb5yXKjXCQV0,4326
-optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py,sha256=dAHV9NgdpXHyTJGT0lieXOB3Pzi_NPlR4rqmRtmAWzM,32412
+optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py,sha256=1umcuZg_ifIyVBfp_6oHQJvZsp_y2UCiv45-rv89_VA,32434
 optimum/rbln/transformers/models/depth_anything/__init__.py,sha256=xvPSIriMJWyNeVYoVB1Z7YqB4kkHOIkaHq7loNps-dk,756
 optimum/rbln/transformers/models/depth_anything/configuration_depth_anything.py,sha256=JujBVEUa_zZDXNPr1y-B_PhK5SgFFcY8Ib4EoGjjtmE,989
 optimum/rbln/transformers/models/depth_anything/modeling_depth_anything.py,sha256=tTmsVaW9Wb2WD3nKRLwp7swn3hbMvgwUEJwwVIfNYEc,1008
@@ -127,10 +127,10 @@ optimum/rbln/transformers/models/gemma/configuration_gemma.py,sha256=H1nVp8HBJxx
 optimum/rbln/transformers/models/gemma/gemma_architecture.py,sha256=4Ry2pFfWg0sVijgTP9SYt1vwJr8DsCVcusg-z0ePX5c,943
 optimum/rbln/transformers/models/gemma/modeling_gemma.py,sha256=AsuFnrwZcRCKtF39BpHHNea0S34N2lNWKV4qZimmY8I,4170
 optimum/rbln/transformers/models/gemma3/__init__.py,sha256=6rugk3615SEt4lh7gduo_J9VyGiSReuEIvL0Uno0eaI,790
-optimum/rbln/transformers/models/gemma3/configuration_gemma3.py,sha256=Hn9WKxTOwciNvBYkoY2y-icQJ6MttMapSyWbNpIxw78,4522
+optimum/rbln/transformers/models/gemma3/configuration_gemma3.py,sha256=NJJfarzbWJc3pm0XvICN7D0FFF9nqidagIEoOvYLixQ,4696
 optimum/rbln/transformers/models/gemma3/gemma3_architecture.py,sha256=fpLDAXCe5paWVsfc0tL59JkRQMRF-WNgIzOIb_QpSLU,6191
 optimum/rbln/transformers/models/gemma3/gemma3_runtime_utils.py,sha256=vYQ9sjRlkfamxZca_hVMQI0ylKeExsV02gOWaYVMjyg,9640
-optimum/rbln/transformers/models/gemma3/modeling_gemma3.py,sha256=7-JlJg9GeMSbtutjSYEY3ZyRJIxz-zqHk7Shshr8Pv4,24228
+optimum/rbln/transformers/models/gemma3/modeling_gemma3.py,sha256=WQ5qRZcM2EYLNqogjy-I04abwZXLZilCGFM4vO-MH4c,25381
 optimum/rbln/transformers/models/gpt2/__init__.py,sha256=SsawHMStE3wYRtqkH5EvdTFkCdX0LLmp-QSKFhEBrHo,740
 optimum/rbln/transformers/models/gpt2/configuration_gpt2.py,sha256=iGdHfzG7plekZcIz-Z5U8lRE4SB8gbJJNcFQJ9l8Myg,1533
 optimum/rbln/transformers/models/gpt2/gpt2_architecture.py,sha256=MyAWReXmyuHnDpW5HI_TI7psyJZxLujZ9KT5XnNm7nA,2802
@@ -140,18 +140,18 @@ optimum/rbln/transformers/models/grounding_dino/configuration_grounding_dino.py,
 optimum/rbln/transformers/models/grounding_dino/grounding_dino_architecture.py,sha256=2BGhyKa7x6fiiZPaLy_S7zKr2NOdJnMLFMf6CEcegGE,26674
 optimum/rbln/transformers/models/grounding_dino/modeling_grounding_dino.py,sha256=bXAOs2QH4sy2UFoFLUSM6u1_VHouUT5COERLQX20F6Y,46897
 optimum/rbln/transformers/models/idefics3/__init__.py,sha256=ulxE7HEfXsNJhd25J9Fvi6vggo9aZH9sLKJjWB6LlzQ,814
-optimum/rbln/transformers/models/idefics3/configuration_idefics3.py,sha256=U3aDw2InB4BIHoujCaK3Q3pzyT_V6RepSdDhBU31WJ0,2773
-optimum/rbln/transformers/models/idefics3/modeling_idefics3.py,sha256=UqKUVZ6pZjP2VMfBa3-dJkLNPDqr3H1wHiOo9LPucjs,19636
+optimum/rbln/transformers/models/idefics3/configuration_idefics3.py,sha256=7IENNxflZL8ZH3YRqtCXfYdKs-RdUeGiPzq-C03te_s,3679
+optimum/rbln/transformers/models/idefics3/modeling_idefics3.py,sha256=ajgLsQsU4n5NwcCqZvG7gswXhjPTHeyTjwjhje5Z7xI,19936
 optimum/rbln/transformers/models/llama/__init__.py,sha256=6tgx9-qlM5r9ouoeZEouVRNLs3r6Sku-cuXNkyfeFHc,746
 optimum/rbln/transformers/models/llama/configuration_llama.py,sha256=_uxfH5kaGbeJTMJfESYn0Vg3OEkINS2ShGtVQTeOcs4,1578
 optimum/rbln/transformers/models/llama/llama_architecture.py,sha256=S7MCPfyjG5eUqgaS-QNBB0ApUD6wnb5fR0RHq7k7-pA,728
 optimum/rbln/transformers/models/llama/modeling_llama.py,sha256=uRxEXYhHOuEwPjBo_Ps3eFU1uwScasla6P8HwsQgAu0,4214
 optimum/rbln/transformers/models/llava/__init__.py,sha256=FaVLgBIqKGjT_nvwYO9k9BVqrzH_Ym3DfjGRCSUhG2s,734
-optimum/rbln/transformers/models/llava/configuration_llava.py,sha256=CmvuNvBjb27FNLPYlZRWfPXFOFP7jltP9ucGK_YAeHA,2544
-optimum/rbln/transformers/models/llava/modeling_llava.py,sha256=3k6lF3k0gjy4zptWH_fiKM7p7kUE1tzhBpwcitm5M1s,17268
+optimum/rbln/transformers/models/llava/configuration_llava.py,sha256=c1rie8LCypxlsT7SNjZJE07_xCLAasV4EBs97o1757Q,2998
+optimum/rbln/transformers/models/llava/modeling_llava.py,sha256=R_wDTJDNVweivDIsZ0eDd5Z21NEcH9O3xdTV9z8mpjQ,20239
 optimum/rbln/transformers/models/llava_next/__init__.py,sha256=kDXKr7wMkp1XqE__DER2B8kQF_NYMxhzsQS5ytGg56I,752
-optimum/rbln/transformers/models/llava_next/configuration_llava_next.py,sha256=cJSncotAJYLMgKUXir7FDVB5avZW_uM65cXtku2pDRw,2759
-optimum/rbln/transformers/models/llava_next/modeling_llava_next.py,sha256=E3w3-kTBzmcG0Xa3LjlSMlO8vFOFhNGoNIWCeQq-_G0,27092
+optimum/rbln/transformers/models/llava_next/configuration_llava_next.py,sha256=Sz8L8p_23T7xw7pkUmW5pyK_wZclph1p_kQYbslc8m8,2708
+optimum/rbln/transformers/models/llava_next/modeling_llava_next.py,sha256=ZGVTOaAOpUOvyVLfSsHpfje4W8FDR_PV6MhS9QNj-Uk,21230
 optimum/rbln/transformers/models/midm/__init__.py,sha256=IC3FETwgYinbp3wDj7tp4zIHJhbqM-c6GfTRdYcMNj8,913
 optimum/rbln/transformers/models/midm/configuration_midm.py,sha256=DxhcSJlApxfi00XxYmSkKZ6bY9vfLXT0zh-oMKkZot0,1365
 optimum/rbln/transformers/models/midm/midm_architecture.py,sha256=RlkmNhaWE5h_awt9aTtR8VZfshNTah0IoUfD2Z9vfxI,5055
@@ -174,7 +174,7 @@ optimum/rbln/transformers/models/phi/modeling_phi.py,sha256=r7B0NlqwIGjm-MmE-h5_
 optimum/rbln/transformers/models/phi/phi_architecture.py,sha256=ygJIJvn20bnxE9nHKo4CBW9_1FJsz7MEVolB5asTmI0,3684
 optimum/rbln/transformers/models/pixtral/__init__.py,sha256=fhclVAWnIDsfMfC-TW6mYrJXxgyehlLaadK64LOShH4,716
 optimum/rbln/transformers/models/pixtral/configuration_pixtral.py,sha256=b79zkJB1jzHx4S1wTe-Ju_Yel_PS5Q8bfmlQPzkchKU,1677
-optimum/rbln/transformers/models/pixtral/modeling_pixtral.py,sha256=4H-Mz8OdDgRUW30N5PnOf6vFqtez2iKgoJIXpdEYvXU,12361
+optimum/rbln/transformers/models/pixtral/modeling_pixtral.py,sha256=P1lzi6JOTB43nBfCOonUDYhIXoMq6DnQpcvGfOO7ZP8,12252
 optimum/rbln/transformers/models/pixtral/pixtral_architecture.py,sha256=s-6C9DtHmSZEGJXo5b95RwZE2A5aR6ELMHlj7aK6CIg,2950
 optimum/rbln/transformers/models/qwen2/__init__.py,sha256=h9dWJ3HX4xspMLt44g7r3UGU8QL03Ynmz_Mi3Vlu6UA,746
 optimum/rbln/transformers/models/qwen2/configuration_qwen2.py,sha256=tTWcPOk_ycZvdSPlal9S5elTmWZAX2BbpZP5Ok2ySwI,1567
@@ -182,12 +182,12 @@ optimum/rbln/transformers/models/qwen2/modeling_qwen2.py,sha256=VOboPJF1rvvSVWkH
 optimum/rbln/transformers/models/qwen2/qwen2_architecture.py,sha256=XlNAMYAcDLohnSAhIFGKOPuCB5XLgzYs5ABWdeQSaZs,720
 optimum/rbln/transformers/models/qwen2_5_vl/__init__.py,sha256=rAW3DKQUzGL6EMwa5r1iLu94yhpiZpk6zfoD7TtYXrc,865
 optimum/rbln/transformers/models/qwen2_5_vl/configuration_qwen2_5_vl.py,sha256=WHLH72i7Pe16Ee1waMixMsR3eD6TsMGN08QD82qdVvw,6162
-optimum/rbln/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py,sha256=hRvA37sPFC9xH1FqnFbtHS9rQOPwAvLYg4zl4oEyK-w,26639
-optimum/rbln/transformers/models/qwen2_5_vl/qwen2_5_vl_architecture.py,sha256=i_UUWhKoFjJ5CCpgeWicqABM23TxMEKPQ354LoZ6iUU,7445
+optimum/rbln/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py,sha256=pI1TNDMO-ZiqXtoHboxLlqfplZbRh22lT1gxhqy6Jtg,26939
+optimum/rbln/transformers/models/qwen2_5_vl/qwen2_5_vl_architecture.py,sha256=y0W5qxrke7JbFNkPTkEOC8TvKnAYZP9bNsQK_IYvxnA,8340
 optimum/rbln/transformers/models/qwen2_vl/__init__.py,sha256=O3t6zKda92CnZDzEnz_dcisMOQ71-OOJxElXzKCH5e0,849
 optimum/rbln/transformers/models/qwen2_vl/configuration_qwen2_vl.py,sha256=mi5CqSKZ77G5Fib3g8a86_4CEB6lb-qJOhDnSqslvNk,4714
-optimum/rbln/transformers/models/qwen2_vl/modeling_qwen2_vl.py,sha256=10NR0LPJFICpPUrINpyMDWVx19HN8sXPpeAiVhE_k2k,20130
-optimum/rbln/transformers/models/qwen2_vl/qwen2_vl_architecture.py,sha256=EZlCuSRTIpSAGEjtDi4SY1V9RRdtgg76ie5jqec1UuI,4833
+optimum/rbln/transformers/models/qwen2_vl/modeling_qwen2_vl.py,sha256=YO8cKBEb7dU9D--gidYsPyhS2arOwgVqDe3tLlGHdx4,20424
+optimum/rbln/transformers/models/qwen2_vl/qwen2_vl_architecture.py,sha256=_5Erjk7udq7YFVVtx4XeWaIzpDlEZKJPu2bBSVBGfKE,5728
 optimum/rbln/transformers/models/qwen3/__init__.py,sha256=tI4KwvXpD35dUUaa8aLUXpWoU9gJGcmKXeywOlH14ZE,746
 optimum/rbln/transformers/models/qwen3/configuration_qwen3.py,sha256=BFRPggnH4VlsXlOa19C6KAID-bPgQ8ooQ29dvogh5zk,2102
 optimum/rbln/transformers/models/qwen3/modeling_qwen3.py,sha256=S05efusxjXJhMMYztstGes6ZbqkSr5I4fHFaLSYVG8c,5760
@@ -204,7 +204,7 @@ optimum/rbln/transformers/models/seq2seq/modeling_seq2seq.py,sha256=G7Rkx4paSDlS
 optimum/rbln/transformers/models/seq2seq/seq2seq_architecture.py,sha256=jmBgj7BkUS_S-T-9DI53rE3KXUHSCoIofr7k5JDVPrU,20024
 optimum/rbln/transformers/models/siglip/__init__.py,sha256=X1Fc1GUnJ2EIxFx45nbeoW-T2t0OyP3W73C0HD8Vowo,712
 optimum/rbln/transformers/models/siglip/configuration_siglip.py,sha256=Fy-ANF91bQno_QVd4ZpyRs-uNgC_XRyBRScBg2uKM6w,3029
-optimum/rbln/transformers/models/siglip/modeling_siglip.py,sha256=1TyRaxmhp6mg6UfhQTbZhW26013TE3nVnroYG7EROcU,8033
+optimum/rbln/transformers/models/siglip/modeling_siglip.py,sha256=9_CeyL9Pgd-ZU7XfqMyoK-mP-ZYL-tia7YyI4wxgKMo,7509
 optimum/rbln/transformers/models/swin/__init__.py,sha256=gUsLDB8ceNxt53Cf69OT32JuZoRdmmIsRfjRdHTLDd0,698
 optimum/rbln/transformers/models/swin/configuration_swin.py,sha256=JE4oMdPhJmRwXxKUWQ3KHccthDLEcDiXEzjMcFx71K0,1690
 optimum/rbln/transformers/models/swin/modeling_swin.py,sha256=npQgTCEkonG41HzHzEk-a13NFLJHA-K82HFW2VyR0xc,13968
@@ -232,6 +232,7 @@ optimum/rbln/transformers/models/xlm_roberta/configuration_xlm_roberta.py,sha256
 optimum/rbln/transformers/models/xlm_roberta/modeling_xlm_roberta.py,sha256=EZd3flRUEE38DYtdqEnG70LV7fHhkamRZV51xrVyjYI,1093
 optimum/rbln/transformers/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 optimum/rbln/transformers/utils/rbln_quantization.py,sha256=pORshQUgTInNaibUtd0HL-T8bKW5wuulZs2q0Oshppc,21659
+optimum/rbln/transformers/utils/rbln_runtime_wrapper.py,sha256=l_-zWpRrp6hp-tDANTrEbspIZH-AUSi_jNJICns_QgE,2672
 optimum/rbln/utils/__init__.py,sha256=ieDBT2VFTt2E0M4v_POLBpuGW9LxSydpb_DuPd6PQqc,712
 optimum/rbln/utils/decorator_utils.py,sha256=xu-TrsNi33SRC2a7DBsyoo6-pEQxWKZPZSmM9QlDe2Y,3745
 optimum/rbln/utils/depreacate_utils.py,sha256=uKxl3ENUCNaZXPnaDQvNxrH8hUIWdBWfZH6BM7ZV__4,385
@@ -241,8 +242,8 @@ optimum/rbln/utils/logging.py,sha256=VKKBmlQSdg6iZCGmAXaWYiW67K84jyp1QJhLQSSjPPE
 optimum/rbln/utils/model_utils.py,sha256=4k5879Kh75m3x_vS4-qOGfqsOiAvc2kdNFFfvsFvz3k,1748
 optimum/rbln/utils/runtime_utils.py,sha256=R6uXDbeJP03-FWdd4vthNe2D4aCra5n12E3WB1ifiGM,7933
 optimum/rbln/utils/save_utils.py,sha256=hG5uOtYmecSXZuGTvCXsTM-SiyZpr5q3InUGCCq_jzQ,3619
-optimum/rbln/utils/submodule.py,sha256=60NGLFvnhjP1DJg1opdb-FVQDsthcLCwWjW_1WQaasU,5280
-optimum_rbln-0.9.1.dist-info/METADATA,sha256=AtptvW3UA4d23uu7EilHiZjQjo2FZOIqmlc7lBsAsho,5348
-optimum_rbln-0.9.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-optimum_rbln-0.9.1.dist-info/licenses/LICENSE,sha256=QwcOLU5TJoTeUhuIXzhdCEEDDvorGiC6-3YTOl4TecE,11356
-optimum_rbln-0.9.1.dist-info/RECORD,,
+optimum/rbln/utils/submodule.py,sha256=SKLnM3KsX8_rv3HauO4oB2-JSjzuadQjRwo_BhMUzLI,6362
+optimum_rbln-0.9.2a0.dist-info/METADATA,sha256=WxK3HKWyo0Ye9NAY7BsxDuPuy0Ykkt4Z90RB7XlCeU8,5350
+optimum_rbln-0.9.2a0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+optimum_rbln-0.9.2a0.dist-info/licenses/LICENSE,sha256=QwcOLU5TJoTeUhuIXzhdCEEDDvorGiC6-3YTOl4TecE,11356
+optimum_rbln-0.9.2a0.dist-info/RECORD,,