optimum-rbln 0.9.3.post1__py3-none-any.whl

optimum/rbln/transformers/models/llama/modeling_llama.py

@@ -0,0 +1,104 @@
+# 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 ....utils import logging
+from ...models.decoderonly import RBLNDecoderOnlyModel, RBLNDecoderOnlyModelForCausalLM
+from .llama_architecture import LlamaWrapper
+
+
+logger = logging.get_logger(__name__)
+
+
+class RBLNLlamaForCausalLM(RBLNDecoderOnlyModelForCausalLM):
+    """
+    The Llama Model transformer with a language modeling head (linear layer) on top.
+    This model inherits from [`RBLNDecoderOnlyModelForCausalLM`]. Check the superclass documentation for the generic methods the library implements for all its models.
+
+    A class to convert and run pre-trained transformers based LlamaForCausalLM model on RBLN devices.
+    It implements the methods to convert a pre-trained transformers LlamaForCausalLM model into a RBLN transformer model by:
+
+    - transferring the checkpoint weights of the original into an optimized RBLN graph,
+    - compiling the resulting graph using the RBLN compiler.
+
+    **Configuration:**
+    This model uses [`RBLNLlamaForCausalLMConfig`] for configuration. When calling methods like `from_pretrained` or `from_model`,
+    the `rbln_config` parameter should be an instance of [`RBLNLlamaForCausalLMConfig`] or a dictionary conforming to its structure.
+
+    See the [`RBLNLlamaForCausalLMConfig`] class for all available configuration options.
+
+    Examples:
+        ```python
+        from optimum.rbln import RBLNLlamaForCausalLM
+
+        # Simple usage using rbln_* arguments
+        # `max_seq_len` is automatically inferred from the model config
+        model = RBLNLlamaForCausalLM.from_pretrained(
+            "meta-llama/Llama-2-7b-hf",
+            export=True,
+            rbln_batch_size=1,
+            rbln_tensor_parallel_size=4,
+        )
+
+
+        # Using a config dictionary
+        rbln_config = {
+            "batch_size": 1,
+            "max_seq_len": 4096,
+            "tensor_parallel_size": 4,
+        }
+        model = RBLNLlamaForCausalLM.from_pretrained(
+            "meta-llama/Llama-2-7b-hf",
+            export=True,
+            rbln_config=rbln_config
+        )
+
+
+        # Using a RBLNLlamaForCausalLMConfig instance (recommended for type checking)
+        from optimum.rbln import RBLNLlamaForCausalLMConfig
+
+        config = RBLNLlamaForCausalLMConfig(
+            batch_size=1,
+            max_seq_len=4096,
+            tensor_parallel_size=4
+        )
+        model = RBLNLlamaForCausalLM.from_pretrained(
+            "meta-llama/Llama-2-7b-hf",
+            export=True,
+            rbln_config=config
+        )
+        ```
+    """
+
+    _decoder_wrapper_cls = LlamaWrapper
+
+
+class RBLNLlamaModel(RBLNDecoderOnlyModel):
+    """
+    The Llama Model transformer outputting raw hidden-states without any specific head on top.
+    This model inherits from [`RBLNDecoderOnlyModel`]. Check the superclass documentation for the generic methods the library implements for all its models.
+
+    A class to convert and run pre-trained transformers based LlamaModel on RBLN devices.
+    It implements the methods to convert a pre-trained transformers LlamaModel into a RBLN transformer model by:
+
+    - transferring the checkpoint weights of the original into an optimized RBLN graph,
+    - compiling the resulting graph using the RBLN compiler.
+
+    **Configuration:**
+    This model uses [`RBLNLlamaModelConfig`] for configuration. When calling methods like `from_pretrained` or `from_model`,
+    the `rbln_config` parameter should be an instance of [`RBLNLlamaModelConfig`] or a dictionary conforming to its structure.
+
+    See the [`RBLNLlamaModelConfig`] class for all available configuration options.
+    """
+
+    _decoder_wrapper_cls = LlamaWrapper

optimum/rbln/transformers/models/llava/__init__.py

@@ -0,0 +1,16 @@
+# 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 .configuration_llava import RBLNLlavaForConditionalGenerationConfig
+from .modeling_llava import RBLNLlavaForConditionalGeneration

optimum/rbln/transformers/models/llava/configuration_llava.py

@@ -0,0 +1,72 @@
+# 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 typing import Any, Optional
+
+from ....configuration_utils import RBLNModelConfig
+from ....utils.logging import get_logger
+
+
+logger = get_logger(__name__)
+
+
+class RBLNLlavaForConditionalGenerationConfig(RBLNModelConfig):
+    """
+    Configuration class for RBLNLlavaForConditionalGenerationConfig.
+
+    This configuration class stores the configuration parameters specific to
+    RBLN-optimized LLaVA models for multimodal conditional generation tasks
+    that combine vision and language processing capabilities.
+    """
+
+    submodules = ["vision_tower", "language_model"]
+
+    def __init__(
+        self,
+        batch_size: Optional[int] = None,
+        vision_tower: Optional[RBLNModelConfig] = None,
+        language_model: Optional[RBLNModelConfig] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            batch_size (Optional[int]): The batch size for inference. Defaults to 1.
+            vision_tower (Optional[RBLNModelConfig]): Configuration for the vision encoder component.
+                This can include settings specific to the vision encoder, such as input resolution or other vision-related parameters.
+                If not provided, default settings will be used.
+            language_model (Optional[RBLNModelConfig]): Configuration for the language model component.
+                This can include settings specific to the language model, such as tensor parallelism or other text-related parameters.
+                If not provided, default settings will be used.
+            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}")
+
+        if self.batch_size != 1:
+            logger.warning("Ignore batch_size for Llava vision tower. It will be set to 1.")
+
+        self.vision_tower = self.initialize_submodule_config(
+            submodule_config=vision_tower,
+            batch_size=1,  # vision_tower batch_size is always 1 in Llava
+            force_kwargs=True,
+        )
+
+        self.language_model = self.initialize_submodule_config(
+            submodule_config=language_model,
+        )

optimum/rbln/transformers/models/llava/modeling_llava.py

@@ -0,0 +1,490 @@
+# 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.
+
+import importlib
+import inspect
+from typing import TYPE_CHECKING, Any, Callable, List, Optional, Tuple, Union
+
+import torch
+from transformers import AutoModelForImageTextToText, LlavaForConditionalGeneration, PretrainedConfig, PreTrainedModel
+from transformers.modeling_outputs import BaseModelOutputWithPooling
+from transformers.modeling_utils import no_init_weights
+from transformers.models.llava.modeling_llava import LlavaCausalLMOutputWithPast
+
+from ....configuration_utils import RBLNCompileConfig, RBLNModelConfig
+from ....modeling import RBLNModel
+from ....utils.logging import get_logger
+from ...modeling_outputs import RBLNDecoderOnlyOutput
+from ...utils.rbln_runtime_wrapper import LoopProcessor
+from ..decoderonly.generation_decoderonly import RBLNDecoderOnlyGenerationMixin
+
+
+logger = get_logger(__name__)
+
+if TYPE_CHECKING:
+    from transformers import AutoFeatureExtractor, AutoProcessor, AutoTokenizer, PretrainedConfig
+
+
+class LoopVisionTower(LoopProcessor):
+    def __init__(self, vision_tower):
+        # FIXME: need to know RBLNModel or RuntimeWrapper
+        if hasattr(vision_tower.model, "runtime"):
+            super().__init__(model=vision_tower)
+        else:
+            super().__init__(model=vision_tower.model[0])
+
+        self.rbln_config = vision_tower.rbln_config
+
+    def _get_batch_size(self, pixel_values, **kwargs):
+        return pixel_values.shape[0]
+
+    def _prepare_inputs_for_iteration(self, index, common_inputs, pixel_values, **kwargs):
+        pixel_values_item = pixel_values[index : index + 1]
+        if "image_sizes" in kwargs and kwargs["image_sizes"] is not None:
+            ret_val = [pixel_values_item, kwargs["image_sizes"][index : index + 1]]
+        else:
+            ret_val = [pixel_values_item]
+
+        out_buffer = [tensor[index : index + 1] for tensor in kwargs["out"]] if "out" in kwargs else None
+        return (ret_val, {"out": out_buffer})
+
+    def _process_outputs(self, outputs: list, **kwargs) -> "BaseModelOutputWithPooling":
+        # when use another Wrapper
+        if hasattr(self.rbln_config, "max_image_size"):
+            last_hidden_states = [output.last_hidden_state for output in outputs]
+            last_hidden_states = torch.cat(last_hidden_states, dim=1)
+            hidden_states = tuple(
+                torch.cat(
+                    [output.hidden_states[layer_idx] for output in outputs],
+                    dim=1,
+                )
+                for layer_idx in range(len(outputs[0].hidden_states))
+            )
+        else:
+            output = kwargs["out"]
+            last_hidden_states = output[0]
+
+            if not output[2:]:
+                hidden_states = None
+            else:
+                hidden_states = tuple(output[2:])
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=last_hidden_states,
+            pooler_output=None,
+            hidden_states=hidden_states,
+        )
+
+
+class LoopProjector(LoopProcessor):
+    def __init__(self, multi_modal_projector: "RBLNModel", rbln_config=None):
+        super().__init__(model=multi_modal_projector)
+        self.rbln_config = rbln_config
+
+    def _get_batch_size(self, image_feature, **kwargs):
+        return image_feature.shape[0]
+
+    def _prepare_inputs_for_iteration(self, index, common_inputs, image_feature, **kwargs):
+        image_feature_item = image_feature[index : index + 1]
+        if hasattr(self.rbln_config.vision_tower, "max_image_size"):
+            out_buffer = [
+                tensor[:, index * image_feature.shape[1] : (index + 1) * image_feature.shape[1], :]
+                for tensor in kwargs["out"]
+            ]
+        else:
+            out_buffer = [tensor[index : index + 1] for tensor in kwargs["out"]]
+        return ([image_feature_item], {"out": out_buffer})
+
+    def _process_outputs(self, outputs: list, **kwargs):
+        output = kwargs["out"]
+        return output[0]
+
+
+class RBLNLlavaForConditionalGeneration(RBLNModel, RBLNDecoderOnlyGenerationMixin):
+    """
+    RBLNLlavaForConditionalGeneration is a multi-modal model that combines vision and language processing capabilities,
+    optimized for RBLN NPUs. It is designed for conditional generation tasks that involve both image and text inputs.
+    This model inherits from [`RBLNModel`]. Check the superclass documentation for the generic methods the library implements for all its models.
+
+    Important Note:
+        This model includes a Large Language Model (LLM) as a submodule. For optimal performance, it is highly recommended to use
+        tensor parallelism for the language model. This can be achieved by using the `rbln_config` parameter in the
+        `from_pretrained` method. Refer to the `from_pretrained` documentation and the RBLNLlavaForConditionalGeneration class for details.
+    Examples:
+        ```python
+        from optimum.rbln import RBLNLlavaForConditionalGeneration
+        model = RBLNLlavaForConditionalGeneration.from_pretrained(
+            "llava-hf/llava-1.5-7b-hf",
+            export=True,
+            rbln_config={
+                "vision_tower": {"output_hidden_states": True},
+                "language_model": {
+                    "tensor_parallel_size": 4,
+                    "use_inputs_embeds": True,  # In Llava, language model must use inputs_embeds as input.
+                },
+            },
+        )
+        model.save_pretrained("compiled-llava-1.5-7b-hf")
+
+        # Using a RBLNLlavaForConditionalGenerationConfig instance (recommended for type checking)
+        from optimum.rbln import RBLNLlavaForConditionalGenerationConfig
+        vision_config = RBLNCLIPVisionModelConfig(
+            batch_size=1,
+            output_hidden_states=True
+        )
+        language_model_config = RBLNLlamaForCausalLMConfig(
+            batch_size=1,
+            max_seq_len=4096,
+            use_inputs_embeds=True,
+            tensor_parallel_size=4
+        )
+        llava_config = RBLNLlavaForConditionalGenerationConfig(
+            batch_size=1,
+            vision_tower=vision_config,
+            language_model=language_model_config
+        )
+        model = RBLNLlavaForConditionalGeneration.from_pretrained(
+            "llava-hf/llava-1.5-7b-hf",
+            export=True,
+            rbln_config=llava_config
+        )
+        ```
+    """
+
+    auto_model_class = AutoModelForImageTextToText
+    _rbln_submodules = [
+        {"name": "vision_tower"},
+        {"name": "language_model"},
+    ]
+
+    def __getattr__(self, __name: str) -> Any:
+        def redirect(func):
+            return lambda *pargs, **kwargs: func(self, *pargs, **kwargs)
+
+        val = getattr(LlavaForConditionalGeneration, __name)
+
+        if isinstance(val, Callable) and "self" in set(inspect.signature(val).parameters):
+            return redirect(val)
+        return val
+
+    def can_generate(self):
+        return True
+
+    @classmethod
+    def _reconstruct_model_if_needed(cls, model: "PreTrainedModel"):
+        with no_init_weights():
+            model_cls_name = model.model.language_model.__class__.__name__
+            causal_model_cls_name = model_cls_name.replace("Model", "ForCausalLM")
+            causal_model_cls = getattr(importlib.import_module("transformers"), causal_model_cls_name)
+            new_language_model = causal_model_cls(model.model.language_model.config)
+
+        new_language_model.lm_head = model.lm_head
+        new_language_model.model = model.model.language_model
+        model.model.language_model = new_language_model
+        model.lm_head = None
+        del model.lm_head
+        return model
+
+    def __post_init__(self, **kwargs):
+        self.vision_tower = LoopVisionTower(self.rbln_submodules[0])
+        self.language_model = self.rbln_submodules[1]
+        self.multi_modal_projector = LoopProjector(self.model[0], rbln_config=self.rbln_config)
+        self.pad_token_id = self.config.pad_token_id if self.config.pad_token_id is not None else -1
+        return super().__post_init__(**kwargs)
+
+    def get_attn_impl(self) -> str:
+        return self.rbln_config.language_model.attn_impl
+
+    def get_kvcache_num_blocks(self) -> int:
+        return self.rbln_config.language_model.kvcache_num_blocks
+
+    def get_input_embeddings(self):
+        return self.language_model.get_input_embeddings()
+
+    @classmethod
+    def _wrap_model_if_needed(cls, model: "PreTrainedModel", rbln_config: RBLNModelConfig):
+        return model.multi_modal_projector
+
+    @classmethod
+    def _update_rbln_config(
+        cls,
+        preprocessors: Optional[Union["AutoFeatureExtractor", "AutoProcessor", "AutoTokenizer"]],
+        model: Optional["PreTrainedModel"] = None,
+        model_config: Optional["PretrainedConfig"] = None,
+        rbln_config: Optional[RBLNModelConfig] = None,
+    ) -> RBLNModelConfig:
+        # support for pixtral that needs padding
+        if hasattr(rbln_config.vision_tower, "max_image_size"):
+            num_positions = (rbln_config.vision_tower.max_image_size[0] // model_config.vision_config.patch_size) * (
+                rbln_config.vision_tower.max_image_size[1] // model_config.vision_config.patch_size
+            )
+            selected_image_feature_dim = num_positions
+
+        else:
+            num_positions = (model_config.vision_config.image_size // model_config.vision_config.patch_size) ** 2 + 1
+            if model_config.vision_feature_select_strategy == "default":
+                selected_image_feature_dim = num_positions - 1
+            else:
+                selected_image_feature_dim = num_positions
+
+        input_info = [
+            (
+                "image_features",
+                [
+                    1,
+                    selected_image_feature_dim,
+                    model_config.vision_config.hidden_size,
+                ],
+                "float32",
+            )
+        ]
+
+        rbln_compile_config = RBLNCompileConfig(input_info=input_info)
+        rbln_config.set_compile_cfgs([rbln_compile_config])
+        return rbln_config
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        inputs_embeds=None,
+        pixel_values=None,
+        attention_mask=None,
+        cache_position=None,
+        image_sizes=None,
+        generate_idx=None,
+        **kwargs,
+    ):
+        is_prefill_phase = generate_idx is None
+        model_inputs = {}
+
+        if is_prefill_phase:
+            generate_idx = attention_mask.sum(dim=-1, keepdim=True).int()
+            cache_position = None
+            pixel_values = pixel_values
+            model_inputs.update({"image_sizes": image_sizes})
+        else:
+            if inputs_embeds is not None:
+                raise NotImplementedError("Specifying inputs_embeds in decoder phase is not supported.")
+
+            pixel_values = None
+            input_ids = input_ids[:, -1:]
+            cache_position = generate_idx
+            generate_idx = generate_idx + 1
+            model_inputs.update({"input_ids": input_ids})
+
+        if inputs_embeds is not None:
+            if self.rbln_config.use_inputs_embeds:
+                model_inputs.update({"inputs_embeds": inputs_embeds})
+            else:
+                raise ValueError(
+                    "The specifying inputs_embeds is only supported when using a compiled RBLN model with 'rbln_use_inputs_embeds' set to True."
+                )
+        else:
+            model_inputs.update({"input_ids": input_ids})
+
+        model_inputs.update(
+            {
+                "attention_mask": attention_mask,
+                "pixel_values": pixel_values,
+                "cache_position": cache_position,
+                "generate_idx": generate_idx,
+            }
+        )
+        return model_inputs
+
+    def _update_model_kwargs_for_generation(self, outputs, model_kwargs, is_encoder_decoder, **kwargs):
+        model_kwargs["generate_idx"] = outputs.generate_idx
+        return model_kwargs
+
+    def get_image_features(
+        self,
+        pixel_values: torch.FloatTensor,
+        vision_feature_layer: Union[int, List[int]],
+        vision_feature_select_strategy: str,
+        **kwargs,
+    ):
+        if vision_feature_select_strategy not in ["default", "full"]:
+            raise ValueError(f"Unexpected select feature strategy: {self.config.vision_feature_select_strategy}")
+
+        kwargs = {k: v for k, v in kwargs.items() if v is not None}
+
+        # prepare out buffer for pre-allocation
+        if hasattr(self.rbln_config.vision_tower, "max_image_size"):
+            vision_out_size = [
+                pixel_values.shape[0],
+                (self.rbln_config.vision_tower.max_image_size[0] // self.config.vision_config.patch_size)
+                * (self.rbln_config.vision_tower.max_image_size[1] // self.config.vision_config.patch_size),
+                self.config.vision_config.hidden_size,
+            ]
+            pooler_out_size = None
+        else:
+            vision_out_size = [
+                pixel_values.shape[0],
+                (self.config.vision_config.image_size // self.config.vision_config.patch_size) ** 2 + 1,
+                self.config.vision_config.hidden_size,
+            ]
+            pooler_out_size = [pixel_values.shape[0], self.config.vision_config.hidden_size]
+
+        vision_out_buffer = []
+        for i in range(self.config.vision_config.num_hidden_layers + 2):
+            vision_out_buffer.append(torch.empty(size=vision_out_size, dtype=torch.float32, device="cpu"))
+        if pooler_out_size is not None:
+            vision_out_buffer.insert(1, torch.empty(size=pooler_out_size, dtype=torch.float32, device="cpu"))
+
+        image_outputs = self.vision_tower(pixel_values, output_hidden_states=True, out=vision_out_buffer, **kwargs)
+
+        if isinstance(vision_feature_layer, int):
+            selected_image_feature = image_outputs.hidden_states[vision_feature_layer]
+            if vision_feature_select_strategy == "default":
+                selected_image_feature = selected_image_feature[:, 1:]
+        else:
+            hs_pool = [image_outputs.hidden_states[layer_idx] for layer_idx in vision_feature_layer]
+            if vision_feature_select_strategy == "default":
+                hs_pool = [hs[:, 1:] for hs in hs_pool]
+            selected_image_feature = torch.cat(hs_pool, dim=-1)
+
+        if hasattr(self.rbln_config.vision_tower, "max_image_size"):
+            num_real_patches = selected_image_feature.shape[1]
+            max_patches = (self.rbln_config.vision_tower.max_image_size[0] // self.config.vision_config.patch_size) * (
+                self.rbln_config.vision_tower.max_image_size[1] // self.config.vision_config.patch_size
+            )
+
+            chunks = []
+            for i in range(0, num_real_patches, max_patches):
+                chunk = selected_image_feature[:, i : i + max_patches, :]
+                chunk_size = chunk.shape[1]
+
+                if chunk_size < max_patches:
+                    padding_tensor = torch.zeros(
+                        (selected_image_feature.shape[0], max_patches - chunk_size, selected_image_feature.shape[2]),
+                        dtype=selected_image_feature.dtype,
+                    )
+                    chunk = torch.cat([chunk, padding_tensor], dim=1)
+                chunks.append(chunk)
+
+            split_features = torch.cat(chunks, dim=0)
+            num_chunks = len(chunks)
+            projector_out_size = [1, max_patches * num_chunks, self.config.text_config.hidden_size]
+            projector_out_buffer = [torch.empty(size=projector_out_size, dtype=torch.float32, device="cpu")]
+            projected_features = self.multi_modal_projector(split_features, out=projector_out_buffer)
+            projected_features = projected_features.view(
+                selected_image_feature.shape[0], num_chunks * max_patches, self.config.text_config.hidden_size
+            )
+            image_features = projected_features[:, :num_real_patches, :]
+        else:
+            projector_out_size = [
+                pixel_values.shape[0] * pixel_values.shape[1],
+                (self.config.vision_config.image_size // self.config.vision_config.patch_size) ** 2,
+                self.config.text_config.hidden_size,
+            ]
+            projector_out_buffer = [torch.empty(size=projector_out_size, dtype=torch.float32, device="cpu")]
+            image_features = self.multi_modal_projector(selected_image_feature, out=projector_out_buffer)
+
+        return image_features
+
+    def _preprocess_prefill(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        pixel_values: Optional[torch.FloatTensor] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        vision_feature_layer: Optional[Union[int, List[int]]] = None,
+        vision_feature_select_strategy: Optional[str] = None,
+        return_dict: Optional[bool] = None,
+        image_sizes: Optional[torch.Tensor] = None,
+        **lm_kwargs,
+    ):
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        vision_feature_layer = (
+            vision_feature_layer if vision_feature_layer is not None else self.config.vision_feature_layer
+        )
+        vision_feature_select_strategy = (
+            vision_feature_select_strategy
+            if vision_feature_select_strategy is not None
+            else self.config.vision_feature_select_strategy
+        )
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if pixel_values is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot specify both pixel_values and inputs_embeds at the same time, and must specify either one"
+            )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.get_input_embeddings()(input_ids)
+
+        if pixel_values is not None:
+            image_features = self.get_image_features(
+                pixel_values=pixel_values,
+                vision_feature_layer=vision_feature_layer,
+                vision_feature_select_strategy=vision_feature_select_strategy,
+                image_sizes=image_sizes,
+            )
+
+            special_image_mask = (input_ids == self.config.image_token_index).unsqueeze(-1)
+            special_image_mask = special_image_mask.expand_as(inputs_embeds)
+            inputs_embeds = inputs_embeds.masked_scatter(special_image_mask, image_features)
+
+        return inputs_embeds
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        pixel_values: Optional[torch.FloatTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        image_sizes: Optional[torch.Tensor] = None,
+        generate_idx: Optional[torch.Tensor] = None,
+        **kwargs,
+    ) -> Union[Tuple, LlavaCausalLMOutputWithPast]:
+        # Prefill
+        if cache_position is None:
+            inputs_embeds = self._preprocess_prefill(
+                input_ids=input_ids, inputs_embeds=inputs_embeds, pixel_values=pixel_values, image_sizes=image_sizes
+            )
+            logits = []
+            inputs = inputs_embeds if inputs_embeds is not None else input_ids
+            batch_size = inputs.shape[0]
+
+            for b_idx in range(batch_size):
+                cache_position = torch.arange(0, generate_idx[b_idx].item(), dtype=torch.int32).unsqueeze(0)
+                output = self.language_model.prefill_decoder(
+                    input_ids=inputs[b_idx : b_idx + 1] if inputs_embeds is None else None,
+                    inputs_embeds=inputs[b_idx : b_idx + 1] if inputs_embeds is not None else None,
+                    attention_mask=attention_mask[b_idx] if attention_mask is not None else None,
+                    cache_position=cache_position,
+                    batch_idx=b_idx,
+                )
+                logits.append(output.logits)
+
+            logits = torch.cat(logits, dim=0)
+
+        # Decoder
+        else:
+            logits = self.language_model.decoder(
+                input_ids=input_ids,
+                inputs_embeds=inputs_embeds,
+                cache_position=cache_position,
+            ).logits
+
+        if not return_dict:
+            return logits, generate_idx
+        else:
+            return RBLNDecoderOnlyOutput(
+                logits=logits,
+                generate_idx=generate_idx,
+            )