optimum-rbln 0.1.7__py3-none-any.whl → 0.1.9__py3-none-any.whl

optimum/rbln/transformers/models/xlm_roberta/modeling_xlm_roberta.py

@@ -0,0 +1,131 @@
+# Copyright 2024 Rebellions Inc.
+
+# 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.
+
+# 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, List, Optional, Union
+
+import torch
+from transformers import AutoModel, PretrainedConfig, PreTrainedModel, XLMRobertaConfig, XLMRobertaModel
+
+from ....modeling_base import RBLNModel
+from ....modeling_config import RBLNConfig, RBLNRuntimeConfig
+
+
+logger = logging.getLogger(__name__)
+
+if TYPE_CHECKING:
+    from transformers import AutoFeatureExtractor, AutoProcessor, AutoTokenizer
+
+
+class RBLNXLMRobertaModel(RBLNModel):
+    auto_model_class = AutoModel  # feature extraction
+    original_model_class = XLMRobertaModel
+    original_config_class = XLMRobertaConfig
+
+    @classmethod
+    def get_pytorch_model(
+        cls,
+        model_id: str,
+        use_auth_token: Optional[Union[bool, str]] = None,
+        revision: Optional[str] = None,
+        force_download: bool = False,
+        cache_dir: Optional[str] = None,
+        subfolder: str = "",
+        local_files_only: bool = False,
+        trust_remote_code: bool = False,
+        rbln_config_kwargs: Optional[Dict[str, Any]] = None,
+        rbln_constructor_kwargs: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ) -> "PreTrainedModel":
+        model: "PreTrainedModel" = super().get_pytorch_model(
+            model_id=model_id,
+            use_auth_token=use_auth_token,
+            revision=revision,
+            force_download=force_download,
+            cache_dir=cache_dir,
+            subfolder=subfolder,
+            local_files_only=local_files_only,
+            trust_remote_code=trust_remote_code,
+            rbln_config_kwargs=rbln_config_kwargs,
+            rbln_constructor_kwargs=rbln_constructor_kwargs,
+            library_name="transformers",
+        )
+
+        return model
+
+    @classmethod
+    def _get_rbln_config(
+        cls,
+        preprocessors: Optional[Union["AutoFeatureExtractor", "AutoProcessor", "AutoTokenizer"]],
+        model_config: Optional["PretrainedConfig"] = None,
+        rbln_max_seq_len: Optional[int] = None,
+        rbln_model_input_names: Optional[List[str]] = None,
+        rbln_batch_size: Optional[int] = None,
+    ) -> RBLNConfig:
+        max_position_embeddings = getattr(model_config, "n_positions", None) or getattr(
+            model_config, "max_position_embeddings", None
+        )
+
+        if rbln_max_seq_len is None:
+            rbln_max_seq_len = max_position_embeddings
+            if rbln_max_seq_len is None:
+                for tokenizer in preprocessors:
+                    if hasattr(tokenizer, "model_max_length"):
+                        rbln_max_seq_len = tokenizer.model_max_length
+                        break
+                if rbln_max_seq_len is None:
+                    raise ValueError("`rbln_max_seq_len` should be specified!")
+
+        if max_position_embeddings is not None and rbln_max_seq_len > max_position_embeddings:
+            raise ValueError("`rbln_enc_max_seq_len` should be less or equal than max_position_embeddings!")
+
+        if rbln_model_input_names is None:
+            # These are BERT's inputs
+            rbln_model_input_names = ["input_ids", "attention_mask", "token_type_ids"]
+
+        if rbln_batch_size is None:
+            rbln_batch_size = 1
+
+        input_info = [
+            (model_input_name, [rbln_batch_size, rbln_max_seq_len], "int64")
+            for model_input_name in rbln_model_input_names
+        ]
+
+        rbln_runtime_config = RBLNRuntimeConfig(input_info=input_info)
+        rbln_runtime_config.batch_size = rbln_batch_size
+
+        meta = {"rbln_max_seq_len": rbln_max_seq_len}
+
+        return RBLNConfig.from_rbln_runtime_configs([rbln_runtime_config], _rbln_meta=meta)
+
+    def forward(
+        self,
+        input_ids: "torch.Tensor",
+        attention_mask: "torch.Tensor",
+        token_type_ids: "torch.Tensor" = None,
+        **kwargs,
+    ):
+        if token_type_ids is None:
+            token_type_ids = torch.zeros_like(input=input_ids, dtype=torch.int64)
+        output = super().forward(input_ids, attention_mask, token_type_ids)
+        return output

optimum/rbln/transformers/utils/rbln_quantization.py

@@ -0,0 +1,109 @@
+# Copyright 2024 Rebellions Inc.
+
+# 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.
+
+# 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 Any, List
+
+import torch
+from torch.nn import Linear, Parameter
+from torch.nn import functional as F
+
+
+QUANTIZED_WEIGHTS = [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "up_proj",
+    "down_proj",
+]
+
+
+def replace_quantized_linear_layers(
+    module: torch.nn.Module,
+) -> None:
+    """Replace target(quantized) linear layer's forward to qlinear forward
+
+    Args:
+        module (torch.nn.Module): The module containing the linear layers to be replaced.
+                                  For example, this could be an instance of a model like
+                                  LlamaForCausalLM().
+    """
+    processed_names: List[str] = []
+
+    for name, layer in module.named_modules():
+        is_replace_linear = name.split(".")[-1] in QUANTIZED_WEIGHTS
+        if isinstance(layer, torch.nn.Linear) and is_replace_linear:
+            *parent_address, child_name = name.split(".")
+            parent = access_attribute(module, parent_address)
+            setattr(parent, child_name, get_qlinear(layer))
+            processed_names.append(name)
+    names_repr = ", ".join(processed_names)
+    print(f"Replace the following linear layers as qlinear layer:\n {{{names_repr}}}")
+
+
+def access_attribute(obj: Any, tokens: List[str]) -> Any:
+    """Get attribute of given object.
+
+    Args:
+        obj: object
+
+        tokens (List[str]): attribute names to access, must be in correct order
+
+    Returns:
+        Any: accessed attribute
+
+    Raises:
+        AttributeError: If attribute doesn't exists
+    """
+    if len(tokens) == 0:
+        return obj
+    return access_attribute(getattr(obj, tokens[0]), tokens[1:])
+
+
+def get_qlinear(layer: Linear):
+    def forward(self, inputs: torch.Tensor) -> torch.Tensor:
+        """Perform weight-only quantized linear layer.
+
+        Forward workflow:
+          - cast weight to high precision
+          - multiply scale factor to weight
+          - call torch.nn.functional linear
+        Note:
+          - Please don't modify following workflow
+          - if the workflow must be changed please contact Rebellions
+        """
+        if inputs.dtype != self.scales.dtype:
+            raise TypeError(f"Expected tensor of dtype {self.scales.dtype} but got {inputs.dtype}")
+        w_fp = self.weight.type(inputs.dtype)
+        w_fp *= self.scales.view(-1, 1)
+        return F.linear(inputs, w_fp, self.bias)
+
+    keep = layer.weight.to(torch.int8)
+    layer.weight = None
+    del layer.weight
+    layer.weight = Parameter(keep, requires_grad=False)
+    layer.scales = Parameter(torch.ones(layer.out_features, dtype=torch.float32), requires_grad=False)
+    layer.forward = lambda *args, **kwargs: forward(layer, *args, **kwargs)
+    return layer

optimum/rbln/utils/import_utils.py

@@ -53,8 +53,7 @@ def is_rbln_available() -> bool:
 
 
 def check_version_compats() -> None:
-    warnings.filterwarnings(action="always", category=ImportWarning)
-
+    warnings.filterwarnings(action="always", category=ImportWarning, module="optimum.rbln")
     my_version = importlib.metadata.version("optimum-rbln")
     target_version = list(filter(lambda v: Version(my_version) > Version(v), RBLN_VERSION_COMPATS.keys()))[0]
     for compat in RBLN_VERSION_COMPATS[target_version]:
@@ -70,5 +69,3 @@ def check_version_compats() -> None:
                 "Please refer to our SDK release notes at https://docs.rbln.ai/about_atom/release_note.html",
                 ImportWarning,
             )
-
-    warnings.resetwarnings()

optimum/rbln/utils/runtime_utils.py

@@ -42,8 +42,9 @@ class RBLNPytorchRuntime:
         return self.forward(*args, **kwds)
 
     def forward(self, *args: List["torch.Tensor"], **kwargs: Dict[str, "torch.Tensor"]):
+        # filtering uselss args or kwarg such as None.
         args = list(filter(lambda arg: isinstance(arg, torch.Tensor), args))
-        kwargs = dict(filter(lambda kwarg: isinstance(kwarg[1], torch.Tensor), kwargs.items()))
+        kwargs = dict(filter(lambda kwarg: isinstance(kwarg[1], torch.Tensor) or kwarg[0] == "out", kwargs.items()))
         output = self.runtime(*args, **kwargs)
         return output
 

{optimum_rbln-0.1.7.dist-info → optimum_rbln-0.1.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: optimum-rbln
-Version: 0.1.7
+Version: 0.1.9
 Summary: Optimum RBLN is the interface between the Hugging Face 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.
 Keywords: transformers,diffusers,inference,rbln,atom,rebel
@@ -21,10 +21,12 @@ Project-URL: Homepage, https://rebellions.ai
 Project-URL: Documentation, https://docs.rbln.ai
 Requires-Python: <3.11,>=3.8
 Requires-Dist: torch<=2.2.1
-Requires-Dist: optimum>=1.17.1
+Requires-Dist: torchvision<=0.17.1
+Requires-Dist: torchaudio<=2.2.1
+Requires-Dist: optimum<=1.20.0
 Requires-Dist: accelerate>=0.28.0
-Requires-Dist: transformers<=4.40.2
-Requires-Dist: diffusers<=0.29.2
+Requires-Dist: transformers<=4.40.2,>=4.38.0
+Requires-Dist: diffusers<=0.30.1
 Requires-Dist: einops>=0.8.0
 Requires-Dist: packaging>=24.1
 Requires-Dist: pytest>=8.1.1; extra == "tests"
@@ -35,7 +37,6 @@ Requires-Dist: sentencepiece>=0.2.0; extra == "tests"
 Requires-Dist: datasets>=2.18.0; extra == "tests"
 Requires-Dist: sacremoses>=0.1.1; extra == "tests"
 Requires-Dist: safetensors>=0.4.2; extra == "tests"
-Requires-Dist: black>=24.3.0; extra == "quality"
 Requires-Dist: ruff>=0.3.3; extra == "quality"
 Requires-Dist: isort>=5.13.2; extra == "quality"
 Requires-Dist: hf-doc-builder>=0.5.0; extra == "quality"
@@ -100,6 +101,11 @@ To install optional dependencies from all groups, specify `-G:all` option.
 pdm install -G:all
 ```
 
+If you want to install optimum-rbln as [editable mode](https://pip.pypa.io/en/stable/topics/local-project-installs/#editable-installs) in existing venv,
+```bash
+(venv) pip install -e .
+```
+
 ## How to use it?
 
 ### Quick Start

{optimum_rbln-0.1.7.dist-info → optimum_rbln-0.1.9.dist-info}/RECORD

@@ -1,17 +1,17 @@
-optimum/rbln/__init__.py,sha256=m2CcYYJw98tSvIFSNJJc5yzrjYKsUdEeBBvsX3a5koI,4251
-optimum/rbln/__version__.py,sha256=V7LnX330m3uiAO0EYQbPUYETPj2br2y1Pv-a7ApMj40,21
-optimum/rbln/diffusers/__init__.py,sha256=JWeu2ihHKiYD0Uzs9jXbaAq-bA1G86UCMPPx_oiJYFU,2606
+optimum/rbln/__init__.py,sha256=Nci6sT3QWzbyKOuJ_KzbIJd8LqkNLS66TAibBUmnPig,4897
+optimum/rbln/__version__.py,sha256=m1D6fscvvsMhq5HVNKw7kP5M8AqEzQm1ekrn_nLQF1M,21
+optimum/rbln/diffusers/__init__.py,sha256=w4W7Wy-Mmh8CQZ5M9JnrrE5bN0UsfNehZI41QadE-hk,2605
 optimum/rbln/diffusers/models/__init__.py,sha256=aY6Llq_31dZjdB9HPBDvi7sXVtdQT9r11gokXG5ffxA,1139
-optimum/rbln/diffusers/models/autoencoder_kl.py,sha256=qIhXCfEADNTm2U9I5ZFN1IfA01zwupUY0IBnJwvxLwI,9506
-optimum/rbln/diffusers/models/controlnet.py,sha256=7T5E-RvGawT2uEtuJYxGTrzIDbApcF13zuXbVCcoQVI,9224
-optimum/rbln/diffusers/models/unet_2d_condition.py,sha256=tdNQHSdN92MlErpsvPpiUleRGhRa9GH0FSFZoSA6-wk,14468
+optimum/rbln/diffusers/models/autoencoder_kl.py,sha256=xfjlbbvNmSrxRGlqNmvuCO9wKaRlcpMF7AxZneitTHM,9520
+optimum/rbln/diffusers/models/controlnet.py,sha256=ePIicWNFKwTBjmH5wDsd1C3LipTHpWpE-X5ZGAMQiDU,9329
+optimum/rbln/diffusers/models/unet_2d_condition.py,sha256=zU38gThPlzs7wAPLDdcojMkmx1P25ooQ29TNEQ034YA,14493
 optimum/rbln/diffusers/pipelines/__init__.py,sha256=Xr_bQbpbC5HbJB2NuUcVQu2BGebDkc2bhsGJmL6jgps,1449
 optimum/rbln/diffusers/pipelines/controlnet/__init__.py,sha256=k0govvSBxBUR5qpxUGxRMHuQCMX7hXHVZ4EqVRw1LWk,1377
-optimum/rbln/diffusers/pipelines/controlnet/multicontrolnet.py,sha256=-Mtot-EJbYnySLKX7v0Im9UQyo2H2HjlZiO31SosbbQ,9592
-optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet.py,sha256=QK8C9mCgCCPTy_sj6PjiwlQzj8sKwUSDWKYeMT3Vb7A,39936
-optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py,sha256=8wUn937eZvr8uVLhTtw-OwF9r_iwdQ1_RpD4XNkarAU,38481
-optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py,sha256=qe2ZW-ugpnfatx7bx1a21C_40kVMSp8DsQ5fl2DFoKM,49849
-optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py,sha256=DKai4mdW_fkhdD4yjIGKyKv00COFH-Nz5ux9jmatjWE,51196
+optimum/rbln/diffusers/pipelines/controlnet/multicontrolnet.py,sha256=rCvQgf5kiqw_b5pfPTpx2GpjoHW-hQsl_4ikYN9klOc,5128
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet.py,sha256=qX0nDFShyZL3RHKgaeeM8XpMLLDsa_PCrhqS2-IfEwM,42605
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py,sha256=9-zw07G5dICByN4k9UgZ5NwPiToRcOApj7M93SM75Pk,41199
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py,sha256=g8SABYroDmcND-0_3CUH1wkdrZPHWdzoWpWLxBk8p-8,53126
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py,sha256=mOlVPumpYbss18jjnZUSwS1EzFjzQDtZf1yOkS4VLng,54485
 optimum/rbln/diffusers/pipelines/stable_diffusion/__init__.py,sha256=qf_uMWSwD-CyRMRC73y1QsTMyl_qCMreIdg0a8rhJuA,1142
 optimum/rbln/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py,sha256=rVryl7riAgkkNJzbXQHKRDYEyR7ZhsF_aF_MkMnerco,5399
 optimum/rbln/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py,sha256=VfOOybXQnotWIS1ch0a-eMSM-BDrPlZdGwtsYtsH0JQ,5747
@@ -19,47 +19,60 @@ optimum/rbln/diffusers/pipelines/stable_diffusion_xl/__init__.py,sha256=8MDMHIVs
 optimum/rbln/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py,sha256=aCJSoXks7IpjX4rcH6q0TjXtIPzNrbvAvz0KbIEmMr8,5684
 optimum/rbln/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py,sha256=Cv9L4El2GOE-3IRQKHNEMuSdWGmtVsRnQJShcv2hOo0,5874
 optimum/rbln/modeling.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-optimum/rbln/modeling_alias.py,sha256=pyYGME31QaiBaLqnjQq3LwUF1T6dLeb8QLB91gzNsLM,1574
-optimum/rbln/modeling_base.py,sha256=779VbQy6gxCkCaa75KAbP87EqBkTQV4RW_roqjxrNMg,29564
+optimum/rbln/modeling_alias.py,sha256=4E0HgaYJTesetqMSnVYKDZ-SRXlTeQ2GZsJ8xuMkmTY,2094
+optimum/rbln/modeling_base.py,sha256=oSm9w1YVCNIQEDW5bdm4bwUF2bqCIGrnrcl335YgVe0,35491
 optimum/rbln/modeling_config.py,sha256=R0GBd-upavJrpR-2SvfgCCxP7f5Zr0NxIqdKmwBfVCk,6439
 optimum/rbln/modeling_seq2seq.py,sha256=XXYu_hpxOgQmMgayUy9URQwgMl9Ci2AfWyVHm7tMP5o,16783
-optimum/rbln/transformers/__init__.py,sha256=fgRvfcQps-YEpXV3h0uz0VBXB1Ts9t-DZRwjS0zHUNU,1992
+optimum/rbln/transformers/__init__.py,sha256=ZC7i3PyMbcZDjxE5F93uZRMyrMghGSmG1ZV3rnAr8A4,2269
+optimum/rbln/transformers/cache_utils.py,sha256=VfArIkAJn3zPXdu-6RXiCWlU-uVwxvhgoMiGxrPtk40,3835
 optimum/rbln/transformers/generation/__init__.py,sha256=6MmqS9D21ir4dcH6_fq8kRsX1VK3QspSn6Qw83F4ORE,1081
 optimum/rbln/transformers/generation/streamers.py,sha256=X-dEmL1L_0Oy0QSFj2RNdamb_xbDWLXd-Ms8ckx6OZ4,5778
-optimum/rbln/transformers/generation/utils.py,sha256=F8gnVYG79kzg_IvQynD-p409E_5loy5VaIXvfi094u0,19464
-optimum/rbln/transformers/models/__init__.py,sha256=8sOIypsvyrKk3bxsdiibCspmQVxc-xMx3WsUbDyMOfM,1319
+optimum/rbln/transformers/generation/utils.py,sha256=XqPbYQCe6cEuWssG9iHpbbL-wgSQTcyKHabGwKj7lnE,19462
+optimum/rbln/transformers/models/__init__.py,sha256=kCbAvlMadrhOv8PyEpvJLLBq1b7DYC3zDmlsxLPufTI,1448
 optimum/rbln/transformers/models/bart/__init__.py,sha256=SGUcpQ_5iLsVxySxtbwhRpmGt7BgVUTxHAjxAjQStdU,1063
-optimum/rbln/transformers/models/bart/bart_architecture.py,sha256=H8yVoBFa5uMXQv_wYCHKRW6tIIjdD50ho9C0vcMsbSo,14956
+optimum/rbln/transformers/models/bart/bart_architecture.py,sha256=T9GjcsL8fAJcvAs_ifnZuDP2F77hhbjBcsc7u53k6OE,14951
 optimum/rbln/transformers/models/clip/__init__.py,sha256=tbco8qW9QhBe3dtWoKgslLZMsXu9dg_KfJ4IgjvK248,1071
-optimum/rbln/transformers/models/clip/modeling_clip.py,sha256=WEXc9LWbClEzxeIw-LliI1A5OpiL7nnDm0G2IzYdqp4,3990
+optimum/rbln/transformers/models/clip/modeling_clip.py,sha256=V8Ix1X9gqeBHXP-HvRuNarFRC-cm--ZEU9ICCvwv-tk,4015
+optimum/rbln/transformers/models/decoderonly/__init__.py,sha256=AG3ib8iZAEDAvVTNhieCyojWZtA67voPB0dI8lbCXTQ,1371
+optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py,sha256=QJHCrYKAyXOHZsKiRt4uRCxoB7zATxjOERSbqOqjK-g,19095
+optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py,sha256=DpY88CqhhesQVh6z42nvzG9p1Ju9wuEloFjPT8F92t4,18436
 optimum/rbln/transformers/models/dpt/__init__.py,sha256=R8OHDxOAYPjkk5t8osaPqRh85Pf1Cg1BtzqesqFRcTI,1045
 optimum/rbln/transformers/models/dpt/modeling_dpt.py,sha256=uN_5DhjGbFmTbpm1JUBgPsDhMP_vIyc0QM2UK5DoRqc,3537
+optimum/rbln/transformers/models/gemma/__init__.py,sha256=L1Qfr6dufWtoUZND_ofwxXPSxivEvPR8exy16a_nM2o,1042
+optimum/rbln/transformers/models/gemma/gemma_architecture.py,sha256=hT0CqL_jhKWi8cDa1zFcAFPyli844wkliJ3bL5OyEdQ,4376
+optimum/rbln/transformers/models/gemma/modeling_gemma.py,sha256=QW57x2ehojt0CHBJSC4Y51VaMqvjbiSYfEpxIPVMbj8,2643
 optimum/rbln/transformers/models/gpt2/__init__.py,sha256=jsOKYXUclG9G6cwUTUX4eeKqjCPfQUwev7TTFIMXS4Y,1040
-optimum/rbln/transformers/models/gpt2/gpt2_architecture.py,sha256=QiNKLhBZ_-1bcq-7WQ4Bd2MK5xj9XR35IdkwEgCA6lk,10004
-optimum/rbln/transformers/models/gpt2/modeling_gpt2.py,sha256=coWguusgbBhQ8yUfl7EFOCPnzeujDQpepRdd09czRZg,12471
+optimum/rbln/transformers/models/gpt2/gpt2_architecture.py,sha256=SIzwfQFRNvNOHjZmKcUZiURbnmXNXFh95twYELVmHcg,10278
+optimum/rbln/transformers/models/gpt2/modeling_gpt2.py,sha256=ZR0wep7V_sENaODlAbg70_xNZKeTK8xdN0AEOsIrKms,2721
 optimum/rbln/transformers/models/llama/__init__.py,sha256=5mX-MuKzVBj6WQeVxyPhtvFTv0jeZXAFfg4RZ2nVUh0,1042
-optimum/rbln/transformers/models/llama/llama_architecture.py,sha256=bBUZBAsBvsvxY-_gGUofm5zn-piu61pooZr86Ib2CuI,27086
-optimum/rbln/transformers/models/llama/llama_architecture_cb.py,sha256=bhbi0MAee0k1cHeB20c0maD89fxT-53DSf1Mn8ZhgjA,32719
-optimum/rbln/transformers/models/llama/modeling_llama.py,sha256=kTMxJwHRdK0XJFu_OfVWq3iWdOXZkDf4TdKzsi0uQWQ,19821
+optimum/rbln/transformers/models/llama/llama_architecture.py,sha256=j4mifSOaIk7wwV9fL9wQSt5kR3rpnvjtxd3VzhMNdgY,1123
+optimum/rbln/transformers/models/llama/modeling_llama.py,sha256=AQU4RVWQb0Ht_eAEiNTkcEq9bPCr-P1JnCkLy77yDnE,2643
 optimum/rbln/transformers/models/midm/__init__.py,sha256=_6kYchy47frGMZ8uoUspZ9IwrmCBQJ-8kVfXM7xOMew,1249
 optimum/rbln/transformers/models/midm/hf_hub_cached/configuration_midm.py,sha256=P5JqTTcx56HOccxKbR14ZjA67BI0RNnJycG738JMaJ4,833
-optimum/rbln/transformers/models/midm/hf_hub_cached/midm_bitext_tokenization.py,sha256=p8U2Owo8KJzOnrI5vAcDkT2DCt3r-05zFDD2m6D4pEg,12835
-optimum/rbln/transformers/models/midm/hf_hub_cached/modeling_midm.py,sha256=v5M_uQsdRUyPaiWEATv_FHp-2Duq2moyQJKSFVY-k1U,61035
+optimum/rbln/transformers/models/midm/hf_hub_cached/midm_bitext_tokenization.py,sha256=5lhMXfqnIak1PJ9YL-vUxIdY_3DUr3IBXzTqf3ofpmI,12835
+optimum/rbln/transformers/models/midm/hf_hub_cached/modeling_midm.py,sha256=54__wd9EXwGxmaHDksTTcUD2aWl6WoszYsR8dlL1wfE,61031
 optimum/rbln/transformers/models/midm/hf_hub_cached/rotary_position_embedding.py,sha256=5ywaUVKTvqO8GRsHOSXOOGlbiEn-DbGkpJs59_dFb18,4059
-optimum/rbln/transformers/models/midm/midm_architecture.py,sha256=G3fSKuh9CGZXyjM1UPZ3wQAYDDLJZcRlKmV_NgcyfJE,19138
-optimum/rbln/transformers/models/midm/modeling_midm.py,sha256=UAZRE9PIVomfA7XgCc1quXl3Kfb2J1rKH-dmSf50EdE,15214
+optimum/rbln/transformers/models/midm/midm_architecture.py,sha256=IFnu54MVPFEk5pvaeJ8RZGlCR7X-9MMeTKThXOP6_M0,11367
+optimum/rbln/transformers/models/midm/modeling_midm.py,sha256=rk1LgR3s9dDC1zwE6Jg6LJQK868VyAIViD9zsK09-UE,2779
+optimum/rbln/transformers/models/mistral/__init__.py,sha256=XtuOmzBITjj-H1yctXobJjHF908x1Wlxr_p4hi06v8I,1046
+optimum/rbln/transformers/models/mistral/mistral_architecture.py,sha256=LCvY4L0Wq1VruKhZ3JTSiuZJqQRJlTae5A2bKsUBGAg,1128
+optimum/rbln/transformers/models/mistral/modeling_mistral.py,sha256=77CRdlD3n465fQRZ6SOya9jRgL7M3KN624VzRUJBxt4,2678
 optimum/rbln/transformers/models/t5/__init__.py,sha256=dK6F1jbBf001h79WZiVdiNZoXm5kOe2fskzhREhu0EE,1057
 optimum/rbln/transformers/models/t5/t5_architecture.py,sha256=2nFovfOdiJdY9jdAR9BngwPO3d2Oofn9jqVWgZ-YYZ0,18091
 optimum/rbln/transformers/models/wav2vec2/__init__.py,sha256=mz4cXqG9b0tDpTAw3qYn3FaJuolX601VmKBE3gohLSw,1043
-optimum/rbln/transformers/models/wav2vec2/modeling_wav2vec2.py,sha256=kooQ1CC6p2mHvRHkFE48d69yNTnkG_V6g9Beu6Sy3XU,4063
+optimum/rbln/transformers/models/wav2vec2/modeling_wav2vec2.py,sha256=ZnsJ9f2Lu3WNH2QFiuOCTU5Y--_wNlyYy_0zuvkJ5RI,4088
 optimum/rbln/transformers/models/whisper/__init__.py,sha256=PZ8qeAAFMas2MizwVYFxlpFWd5k1Pe1x-0IJfYAMhT8,1059
 optimum/rbln/transformers/models/whisper/modeling_whisper.py,sha256=L49ThCv5sqidNevBGsCpGrOSH4H6wzXOCmON1PCmY9M,11996
-optimum/rbln/transformers/models/whisper/whisper_architecture.py,sha256=QX1Nmq26F_82EYgKmdgXEpE2F8ry-inkn2BB9Lx5M38,15885
+optimum/rbln/transformers/models/whisper/whisper_architecture.py,sha256=QtHP5bXUa9HXa95rEdaqsmbH3eG2QJgUBTV1OG7LF-I,15879
+optimum/rbln/transformers/models/xlm_roberta/__init__.py,sha256=NTj4hCpd8L2_i5DZuV5wp-h8OlTLYVUqTrJxzY_Dg9g,1047
+optimum/rbln/transformers/models/xlm_roberta/modeling_xlm_roberta.py,sha256=YOnybpasUwtAhZSRirHQj0kvmzpD0i2nBzcYT0En3ew,5018
+optimum/rbln/transformers/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+optimum/rbln/transformers/utils/rbln_quantization.py,sha256=kRms512Vf1o7vTZl5cd64ivjBVltZrUDCVqCRWFft88,3734
 optimum/rbln/utils/__init__.py,sha256=F6hJP00eV1_hT_IVwqqYwLWcLQAvZbmmrNMJTia3mjI,1106
-optimum/rbln/utils/import_utils.py,sha256=i2GmQJC9kl4BvXncVUrqx8VCqfv1omaHiWyCliBxChg,2632
-optimum/rbln/utils/runtime_utils.py,sha256=EzEabg2E18nq2WZRDZWsZ_hgrdgQ7u_NElTMAYpSDvM,2545
+optimum/rbln/utils/import_utils.py,sha256=btQpDE5WBJUfCMzg5S-cokWbTh_IHYGo1tRNiU16vLU,2624
+optimum/rbln/utils/runtime_utils.py,sha256=6APwOmW04DjdRto5ntKZFTw4CuFd194OcQtImcIQD2U,2621
 optimum/rbln/utils/save_utils.py,sha256=eFIPtmiblCJ3MvtxEPxmAR3iuLEUrzpyzwtVotDauhw,3283
-optimum_rbln-0.1.7.dist-info/METADATA,sha256=5B8Cx1-EWbf1C9VoUUiFJ2iXqIk8e-CExfgKgSZMGwU,4360
-optimum_rbln-0.1.7.dist-info/WHEEL,sha256=rSwsxJWe3vzyR5HCwjWXQruDgschpei4h_giTm0dJVE,90
-optimum_rbln-0.1.7.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-optimum_rbln-0.1.7.dist-info/RECORD,,
+optimum_rbln-0.1.9.dist-info/METADATA,sha256=loqigjmaw3tu7yv9gzydFI-JMJcdr9TTGuJWMQx5F44,4578
+optimum_rbln-0.1.9.dist-info/WHEEL,sha256=rSwsxJWe3vzyR5HCwjWXQruDgschpei4h_giTm0dJVE,90
+optimum_rbln-0.1.9.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+optimum_rbln-0.1.9.dist-info/RECORD,,