optimum-rbln 0.1.12__py3-none-any.whl → 0.1.15__py3-none-any.whl

optimum/rbln/utils/model_utils.py

@@ -0,0 +1,53 @@
+# 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.
+
+
+# Prefix used for RBLN model class names
+RBLN_PREFIX = "RBLN"
+
+
+def convert_hf_to_rbln_model_name(hf_model_name: str):
+    """
+    Convert Hugging Face model name to RBLN model name.
+
+    Args:
+        hf_model_name (str): The Hugging Face model name.
+
+    Returns:
+        str: The corresponding RBLN model name.
+    """
+    return RBLN_PREFIX + hf_model_name
+
+
+def convert_rbln_to_hf_model_name(rbln_model_name: str):
+    """
+    Convert RBLN model name to Hugging Face model name.
+
+    Args:
+        rbln_model_name (str): The RBLN model name.
+
+    Returns:
+        str: The corresponding Hugging Face model name.
+    """
+
+    return rbln_model_name.removeprefix(RBLN_PREFIX)

optimum/rbln/utils/runtime_utils.py

@@ -43,7 +43,7 @@ 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.
+        # filtering useless 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) or kwarg[0] == "out", kwargs.items()))
         output = self.runtime(*args, **kwargs)
@@ -67,7 +67,7 @@ class UnavailableRuntime:
         return iter([self])
 
     def forward(self, *args: List["torch.Tensor"], **kwargs: Dict[str, "torch.Tensor"]):
-        raise RuntimeError("RBLN-Runtime is not created, So it is not available.")
+        raise RuntimeError("The model can't run because the runtime hasn't been created.")
 
     def __repr__(self) -> str:
         return "UnavailableRuntime"
@@ -76,17 +76,17 @@ class UnavailableRuntime:
 class ContextRblnConfig:
     _local = threading.local()
 
-    def __init__(self, device=None, device_map=None, create_runtimes=None, optimze_host_mem=None):
+    def __init__(self, device=None, device_map=None, create_runtimes=None, optimize_host_mem=None):
         self.device = device
         self.device_map = device_map
         self.create_runtimes = create_runtimes
-        self.optimze_host_mem = optimze_host_mem
+        self.optimize_host_mem = optimize_host_mem
 
     def __enter__(self):
         self._local.device = self.device
         self._local.device_map = self.device_map
         self._local.create_runtimes = self.create_runtimes
-        self._local.optimize_host_memory = self.optimze_host_mem
+        self._local.optimize_host_memory = self.optimize_host_mem
         return self
 
     def __exit__(self, exc_type, exc_val, exc_tb):

optimum/rbln/utils/submodule.py

@@ -0,0 +1,114 @@
+import importlib
+from pathlib import Path
+from typing import TYPE_CHECKING, Any, Dict, List
+
+from ..modeling_config import RBLNConfig
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel
+
+    from ..modeling_base import RBLNBaseModel
+
+
+class SubModulesMixin:
+    """
+    _rbln_submodules = [
+        {"name": "vision_tower"},
+        {"name": "language_model"},
+    ]
+    """
+
+    _rbln_submodules: List[Dict[str, Any]] = []
+
+    def __init__(
+        self,
+        *,
+        rbln_submodules: List["RBLNBaseModel"] = [],
+        **kwargs,
+    ) -> None:
+        for submodule_meta, submodule in zip(self._rbln_submodules, rbln_submodules):
+            setattr(self, submodule_meta["name"], submodule)
+
+    @classmethod
+    def _export_submodules_from_model(
+        cls,
+        model: "PreTrainedModel",
+        model_save_dir: str,
+        rbln_kwargs: Dict[str, Any],
+        **kwargs,
+    ) -> List["RBLNBaseModel"]:
+        rbln_submodules = []
+        for submodule in cls._rbln_submodules:
+            submodule_name = submodule["name"]
+            torch_submodule: "PreTrainedModel" = getattr(model, submodule["name"])
+            cls_name = torch_submodule.__class__.__name__
+            submodule_cls: "RBLNBaseModel" = getattr(importlib.import_module("optimum.rbln"), f"RBLN{cls_name}")
+
+            if submodule_name in rbln_kwargs:
+                kwargs["rbln_config"] = rbln_kwargs[submodule_name]
+
+            rbln_submodule = submodule_cls.from_model(
+                model=torch_submodule,
+                subfolder=submodule_name,
+                model_save_dir=model_save_dir,
+                **kwargs,
+            )
+
+            rbln_submodules.append(rbln_submodule)
+
+        return rbln_submodules
+
+    @classmethod
+    def _load_submodules_from_compiled_models(
+        cls,
+        model_save_dir: str,
+        rbln_kwargs: Dict[str, Any],
+        **kwargs,
+    ):
+        rbln_submodules = []
+        for submodule in cls._rbln_submodules:
+            submodule_name = submodule["name"]
+
+            if submodule_name in rbln_kwargs:
+                kwargs["rbln_config"] = rbln_kwargs[submodule_name]
+
+            # Get cls name for call the constructor of the rbln class
+            submodule_rbln_config = RBLNConfig.load(Path(model_save_dir) / submodule_name)
+            submodule_cls_name = submodule_rbln_config.meta["cls"]
+            submodule_cls: "RBLNBaseModel" = getattr(importlib.import_module("optimum.rbln"), submodule_cls_name)
+
+            rbln_submodule = submodule_cls._from_pretrained(
+                model_id=model_save_dir,
+                config=None,
+                subfolder=submodule_name,
+                **kwargs,
+            )
+            rbln_submodules.append(rbln_submodule)
+        return rbln_submodules
+
+    @classmethod
+    def _load_submodules(
+        cls,
+        model_save_dir,
+        rbln_kwargs,
+        model=None,
+        **kwargs,
+    ):
+        # Two ways :
+        # 1. Compile from pytorch object
+        # 2. Load from compiled file
+        if model is not None:
+            return cls._export_submodules_from_model(
+                model=model,
+                model_save_dir=model_save_dir,
+                rbln_kwargs=rbln_kwargs,
+                **kwargs,
+            )
+
+        else:
+            return cls._load_submodules_from_compiled_models(
+                model_save_dir=model_save_dir,
+                rbln_kwargs=rbln_kwargs,
+                **kwargs,
+            )

optimum/rbln/utils/timer_utils.py

@@ -12,11 +12,11 @@ logger = get_logger()
 def rbln_timer(print_name):
     def decorator(function):
         def wrapper(*args, **kwargs):
-            disable = os.getenv("OPTIMUM_RBLN_DISABLE_SPIN", 'False').lower() in ('true', '1', 't')
+            disable = os.getenv("OPTIMUM_RBLN_DISABLE_SPIN", "False").lower() in ("true", "1", "t")
             if disable:
                 logger.info(f"{print_name} ...")
 
-            spinner = Halo(text=f"{print_name} ...", spinner='dots', color='green', enabled=(not disable))
+            spinner = Halo(text=f"{print_name} ...", spinner="dots", color="green", enabled=(not disable))
             spinner.start()
 
             # Start timer

optimum_rbln-0.1.15.dist-info/METADATA

@@ -0,0 +1,106 @@
+Metadata-Version: 2.3
+Name: optimum-rbln
+Version: 0.1.15
+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.
+Project-URL: Homepage, https://rebellions.ai
+Project-URL: Documentation, https://docs.rbln.ai
+Author-email: "Rebellions Inc." <support@rebellions.ai>
+License: Apache
+Keywords: atom,diffusers,inference,rbln,rebel,transformers
+Classifier: Development Status :: 2 - Pre-Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Education
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: <3.13,>=3.9
+Requires-Dist: accelerate>=1.0.1
+Requires-Dist: diffusers<=0.31.0
+Requires-Dist: einops>=0.8.0
+Requires-Dist: halo>=0.0.31
+Requires-Dist: packaging>=24.1
+Requires-Dist: torch<=2.5.1
+Requires-Dist: torchaudio<=2.5.1
+Requires-Dist: torchvision<=0.20.1
+Requires-Dist: transformers==4.45.2
+Description-Content-Type: text/markdown
+
+
+# Optimum RBLN
+
+<div align="center">
+
+<img src="assets/rbln_logo.png" width="60%"/>
+
+[![PyPI version](https://badge.fury.io/py/optimum-rbln.svg)](https://badge.fury.io/py/optimum-rbln)
+[![License](https://img.shields.io/github/license/rebellions-sw/optimum-rbln)](https://github.com/rebellions-sw/optimum-rbln/blob/main/LICENSE)
+
+</div>
+
+🤗 Optimum RBLN provides an interface between Hugging Face libraries ([Transformers](https://huggingface.co/docs/transformers), [Diffusers](https://huggingface.co/docs/diffusers/index)) and RBLN Accelerators, including [ATOM](https://rebellions.ai/rebellions-product/rbln-ca25/) and [REBEL](https://rebellions.ai/rebellions-product/rebel/).
+
+This library enables seamless integration between the Hugging Face ecosystem and RBLN's NPU acceleration through a comprehensive toolkit for model loading and inference across single- and multi-Accelerator environments. While we maintain a list of [officially validated models and tasks](https://docs.rbln.ai/software/optimum/optimum_rbln.html), users can easily adapt other models and tasks with minimal modifications.
+
+## Key Features
+
+🚀 **High Performance Inference**
+- Optimized model execution on RBLN NPUs through RBLN SDK compilation
+- Support for both single-NPU and multi-NPU inference
+- Integrated with RBLN Runtime for optimal performance
+
+🔧 **Easy Integration**
+- Seamless compatibility with Huggingface model hub
+- Drop-in replacement for existing Huggingface pipelines
+- Minimal code changes required for NPU acceleration
+
+
+## Documentation
+
+Check out [the documentation of Optimum RBLN](https://docs.rbln.ai/software/optimum/optimum_rbln.html) for more advanced usage.
+
+## Getting Started
+
+### Install from PyPI
+
+To install the latest release of this package:
+
+- Export environment variables to access to RBLN private PyPI.
+  ```bash
+  export REBEL_PYPI_USERNAME=<username>
+  export REBEL_PYPI_PASSWORD=<password>
+  ```
+
+- Install optimum-rbln package:
+  ```bash
+  pip install --index-url https://pypi.rebellions.in/simple optimum-rbln
+  ```
+
+### Install from source
+
+#### Prerequisites
+
+- Install [uv](https://docs.astral.sh/uv/) (refer [this link](https://docs.astral.sh/uv/getting-started/installation/) for detailed commands)
+
+The below command installs optimum-rbln along with its dependencies.
+
+```bash
+git clone https://github.com/rebellions-sw/optimum-rbln.git
+cd optimum-rbln
+./scripts/uv-sync.sh
+```
+
+If you want to install local rebel-compiler as editable mode in uv environment,
+```bash
+uv pip install -e /path/to/rebel_compiler/python
+```
+
+### Need Help?
+
+- Join our [Developer Community](https://discuss.rebellions.ai/)
+- Contact maintainers at [support@rebellions.ai](mailto:support@rebellions.ai)

optimum_rbln-0.1.15.dist-info/RECORD

@@ -0,0 +1,110 @@
+optimum/rbln/__init__.py,sha256=rjaGo_lPR8m4RwnTYuLTOL15KNRKXbD2EGn7j_STXIg,6895
+optimum/rbln/__version__.py,sha256=ZKlmJ822TJ49YEqc2wCAMbrp81vFvzcFa9OTia84voM,413
+optimum/rbln/modeling.py,sha256=GpTLugUsFx5qTjyENwR7263naVZrMugtoVvWFEaQLzQ,23788
+optimum/rbln/modeling_alias.py,sha256=Z9vGv6ca82_mhbYclxIZ6e8jt-gf07g--k3ljdQvtGo,2128
+optimum/rbln/modeling_base.py,sha256=TPcJ8JhFvWepIrmPuMQp_IKLWlTmvy2Wb99rhoz_YDk,19755
+optimum/rbln/modeling_config.py,sha256=va58Gpbn3rolqKu9y2u3vYVT6kynBGpox_jod6cs-j0,10612
+optimum/rbln/modeling_diffusers.py,sha256=VabNyhVN5s8M_fCx18SkR9hAfJqfXBZwz1m4Sl9Yihg,14138
+optimum/rbln/diffusers/__init__.py,sha256=jad5hGtgXfP6ZZzYI4uBnb1Qbt6TwfEIJjFOtdNzCgc,3187
+optimum/rbln/diffusers/models/__init__.py,sha256=CKgWCqCEPrAc-l5SxKcwu7TadkSGvqkpNqpwrXZVv90,1749
+optimum/rbln/diffusers/models/controlnet.py,sha256=rIYshEXkBqAGh7cOpfu2quffVHNJj9SQ-ATsgQkre5o,10889
+optimum/rbln/diffusers/models/autoencoders/__init__.py,sha256=yc1ABZG3xxzWPDGf0ADEeuSz3Nrq4ZP-CwddQ-VvWCU,1039
+optimum/rbln/diffusers/models/autoencoders/autoencoder_kl.py,sha256=EoI8EQYYS0SA4wurM3I2shs5Z6MA-YEXohoxDg40LrE,8554
+optimum/rbln/diffusers/models/autoencoders/vae.py,sha256=Ys802twDAnNIMVRp-lL3Rhb8Gh-jot0IFCLBm68FrK8,2889
+optimum/rbln/diffusers/models/transformers/__init__.py,sha256=2bVk_6nuqcREOIVyR-4w5ksmdWJqyIV7Wxc5x0dqYO8,1048
+optimum/rbln/diffusers/models/transformers/transformer_sd3.py,sha256=C_50IkKUYnpCbRlsTsX07PVI5RLRgnouQyISvZhlVOg,7684
+optimum/rbln/diffusers/models/unets/__init__.py,sha256=-0PyRbBVBFujd7nBh0Z4NOe3RVOlAWyvWLU8r62dqdo,1049
+optimum/rbln/diffusers/models/unets/unet_2d_condition.py,sha256=p-obj4tVxBkQMS8W1On8oFrLK-TXCo3Zksw1bOBXRPw,14467
+optimum/rbln/diffusers/pipelines/__init__.py,sha256=M6UtFsGnGKYnuHkuJnfyR5WzajAgUidVUNGaqLBS6bM,2862
+optimum/rbln/diffusers/pipelines/controlnet/__init__.py,sha256=k0govvSBxBUR5qpxUGxRMHuQCMX7hXHVZ4EqVRw1LWk,1377
+optimum/rbln/diffusers/pipelines/controlnet/multicontrolnet.py,sha256=OnOvqnNCK1WrnU7JH93GSmEdMRW8Z2__eorYaj-zHAw,4424
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet.py,sha256=xlfDdKxgIK1DvIQAbOipA00LAMni8f7z8urpeOAyEkE,34884
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py,sha256=C9DwjPiNSQanqNNWiOHJvw3yZtL8tR4YaQZwVkC02_o,33369
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py,sha256=QvoS7g3dfZ3b1Fq1taIfRBf9kNpoR0jsv-H97hXNQ4A,44466
+optimum/rbln/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py,sha256=hHkNKnBsSGdHeFjf45zIpum0DE6JRcw_NEYFLYYmxSg,45779
+optimum/rbln/diffusers/pipelines/stable_diffusion/__init__.py,sha256=7lX6f9XiqROsArw1X0lGsW06H0TrWKwvM9Ye16vis9Y,1224
+optimum/rbln/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py,sha256=33LQR11kryvQqfQ0Srl8O8QqRsfJnKYUZAQHHSruKTY,1362
+optimum/rbln/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py,sha256=WJ4pRfmE0DIAQANXoytoeVyKBa750j-f5oKOHLUmoyY,1390
+optimum/rbln/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py,sha256=DaXqP_2oik-L7LitAXtaiMqcOZixOui67BWSU_qoxu4,1397
+optimum/rbln/diffusers/pipelines/stable_diffusion_3/__init__.py,sha256=I1IbI4uo2ZqA2uLbhqW67wW7-8DwqeE-qPGc3BTL7dQ,1233
+optimum/rbln/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py,sha256=Hb4cGwz422jWvFefTfrPMSt7__KEdlrq41OjiSpNtFo,1410
+optimum/rbln/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py,sha256=hQUAMqQn62-aezWlwscyJu6QlXjXByJ-NXnrvYVIpRk,1445
+optimum/rbln/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py,sha256=8PorhGURHd01eob_4pkZhH6azjqk77FS2xZFnBED5Yg,1445
+optimum/rbln/diffusers/pipelines/stable_diffusion_xl/__init__.py,sha256=giP9dJokdRT4-a5wdJqA1qW8os5Zz0huCack8nlcyxs,246
+optimum/rbln/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py,sha256=XeeZqH63No7NPIXKXgv8Vxq_I0-iql4jHJ0sIkbYSvw,1390
+optimum/rbln/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py,sha256=o1WTLCFjsDZAtGZ84cykU0F4Qw_iMP0g7jaPw1xobP0,1418
+optimum/rbln/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py,sha256=aozAYG2mba2G9ITspDVuos-eWy0MoxlwnVtxGgtN7zk,1425
+optimum/rbln/transformers/__init__.py,sha256=H__SYW4dhcFb02euqtfFZx212QZxkmKu4kgd2sBtVAs,3756
+optimum/rbln/transformers/cache_utils.py,sha256=Ak6yJlzkXsu5jQ-kWIgO97GrsLpoCezpEgQoZnvjmec,3796
+optimum/rbln/transformers/modeling_rope_utils.py,sha256=ob8haNW5f0tPq26xd4sTg-rMrBT9M2VDP7wxt-PkvYY,13087
+optimum/rbln/transformers/models/__init__.py,sha256=gffOrFFYG3W8ypxpDiUotU-McvjhytffeuUzunjv4nQ,3971
+optimum/rbln/transformers/models/auto/__init__.py,sha256=zMqaMIdGwuZJq4gLjRC-69M2mGUKrd0aRpmb4Rpm6-g,435
+optimum/rbln/transformers/models/auto/auto_factory.py,sha256=JIFL404RVf6kAONhLeEz6z59tdahcUAyVSU8hdZZe0g,7421
+optimum/rbln/transformers/models/auto/modeling_auto.py,sha256=DDx8ZUpWPtTr8ZNv7cO_dUgJtDLXjax7gdxwc-wkkgQ,4328
+optimum/rbln/transformers/models/bart/__init__.py,sha256=-mrB4kmPpEIVk713yEIRtC57AZ7kZ23g4bsOKcvBFNE,1138
+optimum/rbln/transformers/models/bart/bart_architecture.py,sha256=W6XeumvuKv1i7p4YzKM0NgpO3NCnc1qwGXknZZrPlP0,21298
+optimum/rbln/transformers/models/bart/modeling_bart.py,sha256=xzZZf-yZdUkq4F271Wfd1l-Hnm4jjgf_yy6hjVohxbo,5144
+optimum/rbln/transformers/models/bert/__init__.py,sha256=divBpVNrRAdNAPgnQkGiEZI4oJHCJtLuwdYpMbD3dMM,1034
+optimum/rbln/transformers/models/bert/modeling_bert.py,sha256=akbsBTsGTs7wrxPw120ryZspwYkmHAUrM4A8Kr3COw4,4111
+optimum/rbln/transformers/models/clip/__init__.py,sha256=iXZfPPIztzMDOkY3fbEzx9dCkFKKtWYXCpLGfjEUeZE,1092
+optimum/rbln/transformers/models/clip/modeling_clip.py,sha256=DI_N-bQcA_Kj7NSkv9VPvV1zsN6IscctVczE_2_ZkVM,6089
+optimum/rbln/transformers/models/decoderonly/__init__.py,sha256=ozc0c3XBI3-5VHhGvZ0zcv6TD-kIXpDCqsAvdW3JSaY,1222
+optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py,sha256=ratAjwMF3eiHTaXwPNqHqWzmkGEC5fG37xP4mGJOMI8,36833
+optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py,sha256=esQp936mlHuAjb1A75jepyB3QVXsUR7a5WgVUhV7wJw,26132
+optimum/rbln/transformers/models/dpt/__init__.py,sha256=R8OHDxOAYPjkk5t8osaPqRh85Pf1Cg1BtzqesqFRcTI,1045
+optimum/rbln/transformers/models/dpt/modeling_dpt.py,sha256=Re15veJwAz3NaUv2GfrHyGUblW0Jcd2KLv23GutFp64,3805
+optimum/rbln/transformers/models/exaone/__init__.py,sha256=CuWNwnZrbd_cLU7jDDPqC0kZIqx1ii_BYyQ98CKDag0,1253
+optimum/rbln/transformers/models/exaone/exaone_architecture.py,sha256=sf0GF35u2AtyadR6WLxaau_0-JKusSomtfh0ILQMras,3528
+optimum/rbln/transformers/models/exaone/modeling_exaone.py,sha256=-0VjxpBOQLM3PAmrWCJXkTKQEap577bS_izn-gx8Ew8,2141
+optimum/rbln/transformers/models/gemma/__init__.py,sha256=L1Qfr6dufWtoUZND_ofwxXPSxivEvPR8exy16a_nM2o,1042
+optimum/rbln/transformers/models/gemma/gemma_architecture.py,sha256=4irFBUeL1rEnHl-r5be_doz4QaqDN3jYZMcN1aHMLYo,2317
+optimum/rbln/transformers/models/gemma/modeling_gemma.py,sha256=floBHXgogv3iAPyjhjKzbjFHeH67L3FYUKs_wtRm-gM,1924
+optimum/rbln/transformers/models/gpt2/__init__.py,sha256=jsOKYXUclG9G6cwUTUX4eeKqjCPfQUwev7TTFIMXS4Y,1040
+optimum/rbln/transformers/models/gpt2/gpt2_architecture.py,sha256=w4dAVeubsNkGtcapknwyyQ5VevWPTWETg4M6Y_tZ9UI,3359
+optimum/rbln/transformers/models/gpt2/modeling_gpt2.py,sha256=uz29eh8bLWxsm8pVHwvA-X8FThW2khSO-Rjysp3RoQk,1910
+optimum/rbln/transformers/models/llama/__init__.py,sha256=5mX-MuKzVBj6WQeVxyPhtvFTv0jeZXAFfg4RZ2nVUh0,1042
+optimum/rbln/transformers/models/llama/llama_architecture.py,sha256=j4mifSOaIk7wwV9fL9wQSt5kR3rpnvjtxd3VzhMNdgY,1123
+optimum/rbln/transformers/models/llama/modeling_llama.py,sha256=D9efkBVUr7TaOkAkiN_qrtQC0AyzLK7cb7UbZpo4XwI,1924
+optimum/rbln/transformers/models/llava_next/__init__.py,sha256=3vi2rmTeKBydGRFOtxELhxWixZggFMpGex6xqfMgi-I,1064
+optimum/rbln/transformers/models/llava_next/modeling_llava_next.py,sha256=Rkpso3eQ1tHXpfLdRUayut4X3J9zsXjF2in4UVN1Yhs,25883
+optimum/rbln/transformers/models/midm/__init__.py,sha256=_6kYchy47frGMZ8uoUspZ9IwrmCBQJ-8kVfXM7xOMew,1249
+optimum/rbln/transformers/models/midm/midm_architecture.py,sha256=PwGwHaYCHmJ4DRvlieVK_IzsZgG79_n-Y8kQs5NwT0A,5790
+optimum/rbln/transformers/models/midm/modeling_midm.py,sha256=NB1Ie2GN9Ilisd9CMIbENRiTtUV0pK3eTStbuz0yQBg,2129
+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=i3X3HKGNee2ocEmpxdHMxuq7UAOgUs-QWlq2OizqA4g,1954
+optimum/rbln/transformers/models/phi/__init__.py,sha256=LrGFTUo1oQnsPSTlxJqAJVVNUdUwq4u_Bf60RUgjLz4,1038
+optimum/rbln/transformers/models/phi/modeling_phi.py,sha256=JfpuUB6cign-lqcUoprgq3gbQclZFT9HGV-NYVkSads,1910
+optimum/rbln/transformers/models/phi/phi_architecture.py,sha256=GTmqWF6cJn4rhFTkuYxuzEcVTE-fM4dfpV0Ve1Abi9Q,4440
+optimum/rbln/transformers/models/qwen2/__init__.py,sha256=1PLl1rlF14C6eSk3EZaDfyEHPaC4DZ2vwVlrklTkOYg,1042
+optimum/rbln/transformers/models/qwen2/modeling_qwen2.py,sha256=8ldxWKk85snFX_EViA7kgcgKAZ_QSbmQxhlO4yFvhOA,1924
+optimum/rbln/transformers/models/qwen2/qwen2_architecture.py,sha256=-X9OZ4HUCYDtwKnvidkWzCMPh_Xuu1wj-wRXIsQ9Pjg,1115
+optimum/rbln/transformers/models/seq2seq/__init__.py,sha256=Oa11lBWDNQWahqvDco3JIsZldYS-lO8qjpnaGKSfR00,1045
+optimum/rbln/transformers/models/seq2seq/modeling_seq2seq.py,sha256=GSy_9bVdiZ4kavavsX-UJ62RIQn18_7k8i8PN6G2P9E,16131
+optimum/rbln/transformers/models/t5/__init__.py,sha256=H1ns7mquQDkImSI1KT4oTe4owK4s_n28YqVxmZ31TF0,1133
+optimum/rbln/transformers/models/t5/modeling_t5.py,sha256=jBZDKwWXbuKEHUraU0N7P-XUPxznpCNdaRYs4buJY0Y,7776
+optimum/rbln/transformers/models/t5/t5_architecture.py,sha256=k3ROGNSGGuF1gFNV-LxoFFgfxo7ab5GSQA4GIi5MLsI,21074
+optimum/rbln/transformers/models/wav2vec2/__init__.py,sha256=mz4cXqG9b0tDpTAw3qYn3FaJuolX601VmKBE3gohLSw,1043
+optimum/rbln/transformers/models/wav2vec2/modeling_wav2vec2.py,sha256=vRDKoujGRvMvyAlJVJhj_EHf0OtQWbZHickFLzGjmDI,4231
+optimum/rbln/transformers/models/whisper/__init__.py,sha256=PZ8qeAAFMas2MizwVYFxlpFWd5k1Pe1x-0IJfYAMhT8,1059
+optimum/rbln/transformers/models/whisper/generation_whisper.py,sha256=Kwwskbp48wJxEkFGQLlm0L252rO7tx_YLYmOA-_IPwI,3387
+optimum/rbln/transformers/models/whisper/modeling_whisper.py,sha256=DX4tBJxULJY_UCm1Tw4EiGn1FjZioBjZZbFAB1Uipm4,15355
+optimum/rbln/transformers/models/whisper/whisper_architecture.py,sha256=OQzkGa2WSUn3OVQ1DYVOY49N46QvxO1hdEbQ7Ke-o_c,17203
+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=H5SEtmCAuG6pL1ovl4eLPGZ3tx1IPOilsxKvnbFDN-E,3821
+optimum/rbln/transformers/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+optimum/rbln/transformers/utils/rbln_quantization.py,sha256=-abKYe20hnwx1RtPE2Yz7C5slEKYmKohBSscoRoA2bo,7807
+optimum/rbln/utils/__init__.py,sha256=F6hJP00eV1_hT_IVwqqYwLWcLQAvZbmmrNMJTia3mjI,1106
+optimum/rbln/utils/decorator_utils.py,sha256=KDxCPC6G1Au8nokvTGjo--JyZbaWjOLzfJl_oewJ5oQ,2311
+optimum/rbln/utils/hub.py,sha256=o-nA2I_jnB0S2AX0-q9lSpSNcdP_TeqZbHv84Gxxvi4,4592
+optimum/rbln/utils/import_utils.py,sha256=fk8VIS46sB5zNqinfpmJLINjcJhTRSf-xdvp-g98Aps,4287
+optimum/rbln/utils/logging.py,sha256=xIcLmUQoIJoBj3owkXN5_WQkQljcos6J6KSdX35IApw,2271
+optimum/rbln/utils/model_utils.py,sha256=cnQbNtc2KUAJOcB6rHVwF8RpCNJFOTiCa91AQlUlgMM,1673
+optimum/rbln/utils/runtime_utils.py,sha256=DXzRJKvLkiRYspefJsps5mHDpgQl_skA1BfIADsXPTg,3815
+optimum/rbln/utils/save_utils.py,sha256=eFIPtmiblCJ3MvtxEPxmAR3iuLEUrzpyzwtVotDauhw,3283
+optimum/rbln/utils/submodule.py,sha256=UHizJSL3osA5Jiaarjbvl7AUWlXp4p8Pb_9JZKsaoCI,3472
+optimum/rbln/utils/timer_utils.py,sha256=o6EI-7-pcr3LhvCGJ1HIs1KH17yF2CaNpTsbHHbHmzc,1229
+optimum_rbln-0.1.15.dist-info/METADATA,sha256=4Zxw1eSnrtAUDNrEEkOhMal7Ryh2CQ4niAKs-9I-dbc,4248
+optimum_rbln-0.1.15.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+optimum_rbln-0.1.15.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+optimum_rbln-0.1.15.dist-info/RECORD,,

{optimum_rbln-0.1.12.dist-info → optimum_rbln-0.1.15.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: pdm-backend (2.4.3)
+Generator: hatchling 1.27.0
 Root-Is-Purelib: true
 Tag: py3-none-any

optimum/rbln/transformers/generation/streamers.py

@@ -1,139 +0,0 @@
-# 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 List, Optional
-
-import torch
-from transformers import AutoTokenizer, TextIteratorStreamer
-
-
-class BatchTextIteratorStreamer(TextIteratorStreamer):
-    """
-    Streamer that stores print-ready text in a queue, to be used by a downstream application as an iterator. This is
-    useful for applications that benefit from accessing the generated text in a non-blocking way (e.g., in an interactive
-    Gradio demo).
-
-    This iterator extends TextIteratorStreamer to support batching of text generation. Each put operation appends
-    generated text to a batch, and the end operation finalizes the batch by processing and storing the generated
-    sequences.
-
-    Parameters:
-        batch_size (int):
-            The size of each text generation batch.
-        tokenizer (AutoTokenizer):
-            The tokenizer used to decode the tokens.
-        skip_prompt (bool, optional, default=False):
-            Whether to skip the prompt to `.generate()` or not. Useful, for example, for chatbots.
-        timeout (float, optional):
-            The timeout for the text queue. If `None`, the queue will block indefinitely. Useful to handle exceptions
-            in `.generate()` when it is called in a separate thread.
-        **decode_kwargs (dict, optional):
-            Additional keyword arguments to pass to the tokenizer's `decode` method.
-
-    """
-
-    def __init__(
-        self,
-        batch_size: int,
-        tokenizer: "AutoTokenizer",
-        skip_prompt: bool = False,
-        timeout: Optional[float] = None,
-        **decode_kwargs,
-    ):
-        super().__init__(tokenizer, skip_prompt, timeout, **decode_kwargs)
-        self.batch_size: int = batch_size
-        self.token_cache: List[List[int]] = [[] for _ in range(batch_size)]
-        self.print_len = [0] * batch_size
-        self.blocked = False
-
-    def put(self, value):
-        """
-        Receives tokens, decodes them, and prints them to buffer as soon as they form entire words.
-        """
-
-        if len(value.shape) < 2:
-            value = torch.reshape(value, (self.batch_size, value.shape[0] // self.batch_size))
-
-        if self.skip_prompt and self.next_tokens_are_prompt:
-            self.next_tokens_are_prompt = False
-            return
-
-        batch_printable_text = []
-        for i in range(self.batch_size):
-            # Add the new token to the cache and decodes the entire thing
-            self.token_cache[i].extend(value[i].tolist())
-            text = self.tokenizer.decode(self.token_cache[i], **self.decode_kwargs)
-
-            # After the symbol for a new line, we flush the cache.
-            if text.endswith("\n"):
-                printable_text = text[self.print_len[i] :]
-                self.token_cache[i] = []
-                self.print_len[i] = 0
-            # If the last token is a CJK character, we print the characters.
-            elif len(text) > 0 and self._is_chinese_char(ord(text[-1])):
-                printable_text = text[self.print_len[i] :]
-                self.print_len[i] += len(printable_text)
-            # Otherwise, prints until the last space char (simple heuristic to avoid printing incomplete words,
-            # which may change with the subsequent token -- there are probably smarter ways to do this!)
-            else:
-                printable_text = text[self.print_len[i] : text.rfind(" ") + 1]
-                self.print_len[i] += len(printable_text)
-            batch_printable_text.append(printable_text)
-
-        self.on_finalized_text(batch_printable_text)
-
-    def end(self):
-        """Flushes any remaining cache and prints a newline to stdout."""
-        batch_printable_text = []
-        for idx in range(self.batch_size):
-            if len(self.token_cache[idx]) > 0:
-                text = self.tokenizer.decode(self.token_cache[idx], **self.decode_kwargs)
-                printable_text = text[self.print_len[idx] :]
-                self.token_cache[idx] = []
-                self.print_len[idx] = 0
-            else:
-                printable_text = ""
-            batch_printable_text.append(printable_text)
-
-        self.next_tokens_are_prompt = True
-        self.on_finalized_text(batch_printable_text, stream_end=True)
-        self.blocked = False
-
-    def on_finalized_text(self, texts: List[str], stream_end: bool = False):
-        self.text_queue.put(texts, timeout=self.timeout)
-        if stream_end:
-            self.text_queue.put(self.stop_signal, timeout=self.timeout)
-
-    # thkim change for demo
-    def __next__(self):
-        value = self.text_queue.get(timeout=self.timeout)
-        if value == self.stop_signal:
-            raise StopIteration()
-        else:
-            return value
-
-    def block(self):
-        self.blocked = True
-
-    def is_blocked(self):
-        return self.blocked