optimum-rbln 0.2.1a0__py3-none-any.whl → 0.2.1a2__py3-none-any.whl

optimum/rbln/transformers/models/bart/bart_architecture.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from typing import Tuple
 
 import torch

optimum/rbln/transformers/models/bart/modeling_bart.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import inspect
 from typing import TYPE_CHECKING, Any, Callable, Dict, Optional, Union
 
@@ -67,23 +58,33 @@ class RBLNBartModel(RBLNModel):
         if max_position_embeddings is not None and rbln_max_seq_len > max_position_embeddings:
             raise ValueError("`rbln_max_seq_len` should be less or equal than max_position_embeddings!")
 
+        signature_params = inspect.signature(cls.hf_class.forward).parameters.keys()
+
         if rbln_model_input_names is None:
             for tokenizer in preprocessors:
                 if hasattr(tokenizer, "model_input_names"):
-                    rbln_model_input_names = tokenizer.model_input_names
+                    rbln_model_input_names = [name for name in signature_params if name in tokenizer.model_input_names]
                     # BartModel's forward() does not take token_type_ids as input.
                     # (Added because some of the tokenizers includes 'token_type_ids')
                     if "token_type_ids" in rbln_model_input_names:
                         rbln_model_input_names.remove("token_type_ids")
+
+                    invalid_params = set(rbln_model_input_names) - set(signature_params)
+                    if invalid_params:
+                        raise ValueError(f"Invalid model input names: {invalid_params}")
                     break
             if rbln_model_input_names is None and hasattr(cls, "rbln_model_input_names"):
                 rbln_model_input_names = cls.rbln_model_input_names
             elif rbln_model_input_names is None and hasattr(cls, "rbln_model_input_names") is False:
-                input_names_order = inspect.signature(cls.hf_class.forward).parameters.keys()
                 raise ValueError(
                     "Specify the model input names obtained by the tokenizer via `rbln_model_input_names`, "
-                    f"and be sure to make the order of the inputs same as BartModel forward() arguments like ({list(input_names_order)})"
+                    f"and be sure to make the order of the inputs same as BartModel forward() arguments like ({list(signature_params)})"
                 )
+        else:
+            invalid_params = set(rbln_model_input_names) - set(signature_params)
+            if invalid_params:
+                raise ValueError(f"Invalid model input names: {invalid_params}")
+            rbln_model_input_names = [name for name in signature_params if name in rbln_model_input_names]
 
         if rbln_batch_size is None:
             rbln_batch_size = 1

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

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,13 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
-from .modeling_bert import RBLNBertModel
+from .modeling_bert import RBLNBertForMaskedLM, RBLNBertForQuestionAnswering, RBLNBertModel

optimum/rbln/transformers/models/bert/modeling_bert.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import inspect
 import logging
 from typing import TYPE_CHECKING, Any, Dict, Optional, Union
@@ -29,6 +20,7 @@ from transformers import PretrainedConfig
 
 from ....modeling import RBLNModel
 from ....modeling_config import RBLNCompileConfig, RBLNConfig
+from ...modeling_generic import RBLNModelForMaskedLM, RBLNModelForQuestionAnswering
 
 
 logger = logging.getLogger(__name__)
@@ -64,19 +56,29 @@ class RBLNBertModel(RBLNModel):
         if max_position_embeddings is not None and rbln_max_seq_len > max_position_embeddings:
             raise ValueError("`rbln_max_seq_len` should be less or equal than max_position_embeddings!")
 
+        signature_params = inspect.signature(cls.hf_class.forward).parameters.keys()
+
         if rbln_model_input_names is None:
             for tokenizer in preprocessors:
                 if hasattr(tokenizer, "model_input_names"):
-                    rbln_model_input_names = tokenizer.model_input_names
+                    rbln_model_input_names = [name for name in signature_params if name in tokenizer.model_input_names]
+
+                    invalid_params = set(rbln_model_input_names) - set(signature_params)
+                    if invalid_params:
+                        raise ValueError(f"Invalid model input names: {invalid_params}")
                     break
             if rbln_model_input_names is None and hasattr(cls, "rbln_model_input_names"):
                 rbln_model_input_names = cls.rbln_model_input_names
             elif rbln_model_input_names is None and hasattr(cls, "rbln_model_input_names") is False:
-                input_names_order = inspect.signature(cls.hf_class.forward).parameters.keys()
                 raise ValueError(
                     "Specify the model input names obtained by the tokenizer via `rbln_model_input_names`, "
-                    f"and be sure to make the order of the inputs same as BertModel forward() arguments like ({list(input_names_order)})"
+                    f"and be sure to make the order of the inputs same as BertModel forward() arguments like ({list(signature_params)})"
                 )
+        else:
+            invalid_params = set(rbln_model_input_names) - set(signature_params)
+            if invalid_params:
+                raise ValueError(f"Invalid model input names: {invalid_params}")
+            rbln_model_input_names = [name for name in signature_params if name in rbln_model_input_names]
 
         if rbln_batch_size is None:
             rbln_batch_size = 1
@@ -96,3 +98,11 @@ class RBLNBertModel(RBLNModel):
 
         rbln_config.model_cfg.update({"max_seq_len": rbln_max_seq_len})
         return rbln_config
+
+
+class RBLNBertForMaskedLM(RBLNModelForMaskedLM):
+    rbln_model_input_names = ["input_ids", "attention_mask", "token_type_ids"]
+
+
+class RBLNBertForQuestionAnswering(RBLNModelForQuestionAnswering):
+    rbln_model_input_names = ["input_ids", "attention_mask", "token_type_ids"]

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

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,13 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from .modeling_clip import RBLNCLIPTextModel, RBLNCLIPTextModelWithProjection, RBLNCLIPVisionModel

optimum/rbln/transformers/models/clip/modeling_clip.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import logging
 from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, Union
 

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

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,13 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from .modeling_decoderonly import RBLNDecoderOnlyModelForCausalLM

optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import math
 from typing import List, Optional, Tuple
 
@@ -553,15 +544,19 @@ class DecoderOnlyAttention(nn.Module):
         super().__init__()
         self._original_mod = self_attn
         self.layer_idx = self_attn.layer_idx
-        self.num_heads = self._original_mod.num_heads
+        self.num_heads = getattr(self._original_mod, "num_heads", None) or getattr(
+            self._original_mod.config, "num_attention_heads"
+        )
         self.head_dim = self._original_mod.head_dim
         self._phase = "prefill"
         self.scale = torch.tensor(self.get_attn_scale())
 
         if hasattr(self._original_mod, "num_key_value_heads"):
             self.num_key_value_heads = self._original_mod.num_key_value_heads
+        elif hasattr(self._original_mod, "config") and hasattr(self._original_mod.config, "num_key_value_heads"):
+            self.num_key_value_heads = self._original_mod.config.num_key_value_heads
         else:
-            self.num_key_value_heads = self._original_mod.num_heads
+            self.num_key_value_heads = self.num_heads
 
         self.attention = self.get_attention()
         self.__post_init__()
@@ -631,33 +626,21 @@ class DecoderOnlyAttention(nn.Module):
         if batch_size > 1 and self.phase == "prefill":
             raise NotImplementedError(f"batch size should be 1 if prefill phase, but got {batch_size}.")
 
-        # TODO(jongho): flash attn legacy. (clone)
-        _seq_positions = seq_positions.clone().unsqueeze(1)
-
-        _key_states = []
-        _value_states = []
-        _attn_outputs = []
-        for b in range(batch_size):
-            seq_position = _seq_positions[b][0]
-            attn_output, key_state, value_state = self.attention(
-                query_states[b].unsqueeze(0),
-                key_states[b].unsqueeze(0),
-                value_states[b].unsqueeze(0),
-                attention_mask[b].unsqueeze(0) if self.phase == "decode" else attention_mask,
-                past_key_state=past_key_values[self.layer_idx][0],
-                past_value_state=past_key_values[self.layer_idx][1],
-                batch_position=b if self.phase == "decode" else batch_position,
-                seq_position=seq_position,
-                scale=self.scale,
-            )
-            _key_states.append(key_state)
-            _value_states.append(value_state)
-            _attn_outputs.append(attn_output)
-        key_states = torch.cat(_key_states, dim=0)
-        value_states = torch.cat(_value_states, dim=0)
-        attn_outputs = torch.cat(_attn_outputs, dim=0)
-
-        attn_outputs = self.o_proj(attn_outputs)
+        attn_output, key_state, value_state = self.attention(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            past_key_state=past_key_values[self.layer_idx][0],
+            past_value_state=past_key_values[self.layer_idx][1],
+            batch_position=None if self.phase == "decode" else batch_position,
+            seq_position=seq_positions,
+            scale=self.scale,
+        )
+        key_states = key_state
+        value_states = value_state
+
+        attn_outputs = self.o_proj(attn_output)
         past_key_values[self.layer_idx] = key_states, value_states
         return attn_outputs, past_key_values
 
@@ -703,8 +686,13 @@ class AttentionOp(nn.Module):
         value_state = value_state.unsqueeze(2)
         attn_mask = attn_mask.unsqueeze(2)
 
+        if self.phase == "decode":
+            batch_size = key_state.shape[0]
+        else:
+            batch_size = 1
+
         query_state = query_state.view(
-            1,
+            batch_size,
             self.num_key_value_heads,
             self.num_heads // self.num_key_value_heads,
             -1,  # seq len
@@ -736,9 +724,9 @@ class AttentionOp(nn.Module):
                 scale,
             )
 
-        attn_output = attn_output.view(1, self.num_heads, -1, self.head_dim)
+        attn_output = attn_output.view(batch_size, self.num_heads, -1, self.head_dim)
         attn_output = attn_output.transpose(1, 2).contiguous()
-        attn_output = attn_output.reshape(1, -1, self.num_heads * self.head_dim)
+        attn_output = attn_output.reshape(batch_size, -1, self.num_heads * self.head_dim)
 
         return attn_output, key_state.squeeze(2), value_state.squeeze(2)
 
@@ -867,31 +855,23 @@ class DecoderOnlyFlashAttention(DecoderOnlyAttention):
         if cos is not None and sin is not None:
             query_states, key_states = self.apply_rotary_pos_embed(query_states, key_states, cos, sin)
 
-        _key_states = []
-        _value_states = []
-        _attn_outputs = []
-        for b in range(batch_size):
-            seq_position = seq_positions[b][0]  # FIXME: Remove take-take pattern matching
-            attn_output, key_state, value_state = self.attention(
-                query_states[b].unsqueeze(0),
-                key_states[b].unsqueeze(0),
-                value_states[b].unsqueeze(0),
-                attention_mask[b].unsqueeze(0) if self.phase == "decode" else attention_mask,
-                past_key_state=past_key_values[self.layer_idx][0],
-                past_value_state=past_key_values[self.layer_idx][1],
-                batch_position=b if self.phase == "decode" else batch_position,
-                seq_position=seq_position,
-                scale=self.scale,
-            )
-            _key_states.append(key_state)
-            _value_states.append(value_state)
-            _attn_outputs.append(attn_output)
-        key_states = torch.cat(_key_states, dim=0)
-        value_states = torch.cat(_value_states, dim=0)
-        attn_outputs = torch.cat(_attn_outputs, dim=0)
-
-        attn_outputs = self.o_proj(attn_outputs)
+        attn_output, key_state, value_state = self.attention(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            past_key_state=past_key_values[self.layer_idx][0],
+            past_value_state=past_key_values[self.layer_idx][1],
+            batch_position=None if self.phase == "decode" else batch_position,
+            seq_position=seq_positions,
+            scale=self.scale,
+        )
+        key_states = key_state
+        value_states = value_state
+
+        attn_outputs = self.o_proj(attn_output)
         past_key_values[self.layer_idx] = key_states, value_states
+
         return attn_outputs, past_key_values
 
 
@@ -917,8 +897,13 @@ class FlashAttentionOp(AttentionOp):
         value_state = value_state.unsqueeze(2)
         attn_mask = attn_mask.unsqueeze(2)
 
+        if self.phase == "decode":
+            batch_size = key_state.shape[0]
+        else:
+            batch_size = 1
+
         query_state = query_state.view(
-            1,
+            batch_size,
             self.num_key_value_heads,
             self.num_heads // self.num_key_value_heads,
             -1,  # seq len
@@ -952,8 +937,8 @@ class FlashAttentionOp(AttentionOp):
             )
 
         # reshape for removing repeat_kv
-        attn_output = attn_output.view(1, self.num_heads, -1, self.head_dim)
+        attn_output = attn_output.view(batch_size, self.num_heads, -1, self.head_dim)
         attn_output = attn_output.transpose(1, 2).contiguous()
-        attn_output = attn_output.reshape(1, -1, self.num_heads * self.head_dim)
+        attn_output = attn_output.reshape(batch_size, -1, self.num_heads * self.head_dim)
 
         return attn_output, key_state, value_state

optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import inspect
 from dataclasses import dataclass
 from pathlib import Path
@@ -218,14 +209,16 @@ class RBLNDecoderOnlyModelForCausalLM(RBLNModel):
 
     @classmethod
     def get_pytorch_model(cls, *args, **kwargs) -> "PreTrainedModel":
+        logger.debug("Loading the LLM model to the CPU.")  # TODO(jongho): Remove.
+
         rbln_kwargs = kwargs.get("rbln_kwargs", {})
         rbln_quantization = rbln_kwargs.get("quantization", None)
-
         if rbln_quantization is not None and rbln_quantization["format"] == "rbln":
             model = cls.get_quantized_model(*args, **kwargs)
         else:
             model = super().get_pytorch_model(*args, **kwargs)
 
+        logger.debug("Loaded the LLM model to the CPU.")
         return model
 
     @classmethod
@@ -297,8 +290,15 @@ class RBLNDecoderOnlyModelForCausalLM(RBLNModel):
         rbln_attn_impl = rbln_kwargs.get("attn_impl", None)
         rbln_kvcache_partition_len = rbln_kwargs.get("kvcache_partition_len", None)
         rbln_quantization = QuantizationManager.validate_quantization_config(rbln_kwargs.get("quantization", None))
+        rbln_prefill_chunk_size = rbln_kwargs.get("prefill_chunk_size", None)
+
+        if rbln_prefill_chunk_size is None:
+            rbln_prefill_chunk_size = 128
+        elif rbln_prefill_chunk_size % 64 != 0 or rbln_prefill_chunk_size == 0:
+            raise ValueError(
+                f"Invalid rbln_prefill_chunk_size: {rbln_prefill_chunk_size}. It must be a nonzero multiple of 64."
+            )
 
-        prefill_chunk_size = 128
         if rbln_max_seq_len is None:
             rbln_max_seq_len = getattr(model_config, "max_position_embeddings", None) or getattr(
                 model_config, "n_positions", None
@@ -369,7 +369,7 @@ class RBLNDecoderOnlyModelForCausalLM(RBLNModel):
 
         prefill_input_info = get_input_info(
             batch_size=1,
-            query_length=prefill_chunk_size,
+            query_length=rbln_prefill_chunk_size,
             use_inputs_embeds=rbln_use_inputs_embeds,
             hidden_size=hidden_size,
         )
@@ -393,7 +393,7 @@ class RBLNDecoderOnlyModelForCausalLM(RBLNModel):
             {
                 "max_seq_len": rbln_max_seq_len,
                 "batch_size": rbln_batch_size,
-                "prefill_chunk_size": prefill_chunk_size,
+                "prefill_chunk_size": rbln_prefill_chunk_size,
                 "use_inputs_embeds": rbln_use_inputs_embeds,
                 "kvcache_partition_len": rbln_kvcache_partition_len,
                 "attn_impl": rbln_attn_impl,

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

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,13 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from .modeling_dpt import RBLNDPTForDepthEstimation

optimum/rbln/transformers/models/dpt/modeling_dpt.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import logging
 from typing import TYPE_CHECKING, Any, Dict, Iterable, Optional, Union
 

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

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 import os
 from os import environ
 

optimum/rbln/transformers/models/exaone/exaone_architecture.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from typing import TYPE_CHECKING
 
 import torch.nn as nn

optimum/rbln/transformers/models/exaone/modeling_exaone.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 
 from transformers import AutoModelForCausalLM
 

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

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,13 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from .modeling_gemma import RBLNGemmaForCausalLM

optimum/rbln/transformers/models/gemma/gemma_architecture.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Rebellions Inc.
+# 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.
@@ -12,15 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Portions of this software are licensed under the Apache License,
-# Version 2.0. See the NOTICE file distributed with this work for
-# additional information regarding copyright ownership.
-
-# All other portions of this software, including proprietary code,
-# are the intellectual property of Rebellions Inc. and may not be
-# copied, modified, or distributed without prior written permission
-# from Rebellions Inc.
-
 from typing import TYPE_CHECKING
 
 from ...models.decoderonly.decoderonly_architecture import (