From a56612dbcea1e7534d5f7177f5f8523a96ee6cf7 Mon Sep 17 00:00:00 2001
From: ali-12388 <1940747290@qq.com>
Date: Mon, 3 Nov 2025 18:44:33 +0800
Subject: [PATCH 1/6] add qwen3-omni

---
 angelslim/data/__init__.py                    |   1 +
 angelslim/data/dataloader.py                  |  12 ++
 angelslim/data/omni_dataset.py                | 127 +++++++++++++++
 angelslim/engine.py                           |  17 +-
 angelslim/models/__init__.py                  |   1 +
 angelslim/models/model_factory.py             |   4 +-
 angelslim/models/omni/__init__.py             |  16 ++
 angelslim/models/omni/qwen_omni.py            | 149 ++++++++++++++++++
 angelslim/utils/config_parser.py              |   7 +
 .../fp8_dynamic/qwen3_omn_fp8_dynamic.yaml    |  23 +++
 .../fp8_static/qwen3_omni_fp8_static.yaml     |  32 ++++
 dataset/omni_fake_data/fake_data.json         |   3 +
 dataset/omni_fake_data/images/0.png           | Bin 0 -> 44922 bytes
 requirements/requirements.txt                 |   1 +
 tools/run.py                                  |   4 +
 15 files changed, 394 insertions(+), 3 deletions(-)
 create mode 100644 angelslim/data/omni_dataset.py
 create mode 100644 angelslim/models/omni/__init__.py
 create mode 100644 angelslim/models/omni/qwen_omni.py
 create mode 100644 configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml
 create mode 100644 configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
 create mode 100755 dataset/omni_fake_data/fake_data.json
 create mode 100755 dataset/omni_fake_data/images/0.png

diff --git a/angelslim/data/__init__.py b/angelslim/data/__init__.py
index 97d05f98..7e5c29c5 100644
--- a/angelslim/data/__init__.py
+++ b/angelslim/data/__init__.py
@@ -6,5 +6,6 @@
 
 from .dataloader import DataLoaderFactory  # noqa: F401
 from .multimodal_dataset import MultiModalDataset  # noqa: F401
+from .omni_dataset import OmniDataset  # noqa: F401
 from .text2image_dataset import Text2ImageDataset  # noqa: F401
 from .text_dataset import TextDataset  # noqa: F401
diff --git a/angelslim/data/dataloader.py b/angelslim/data/dataloader.py
index 1c92cddb..c41b755d 100644
--- a/angelslim/data/dataloader.py
+++ b/angelslim/data/dataloader.py
@@ -20,6 +20,7 @@
 
 from .base_dataset import BaseDataset
 from .multimodal_dataset import MultiModalDataset
+from .omni_dataset import OmniDataset
 from .text2image_dataset import Text2ImageDataset
 from .text_dataset import TextDataset
 
@@ -39,6 +40,7 @@ def create_data_loader(
         data_type: str = "auto",
         num_workers: int = 0,
         inference_settings: Dict = None,
+        use_audio_in_video: bool = False,
         model_name: str = None,
     ) -> DataLoader:
         """
@@ -98,6 +100,16 @@ def create_data_loader(
                 num_samples=num_samples,
                 inference_settings=inference_settings,
             )
+        elif data_type == "OmniDataset":
+            dataset = OmniDataset(
+                processor=processor,
+                device=device,
+                max_length=max_length,
+                num_samples=num_samples,
+                data_source=data_source,
+                is_hf_dataset=not os.path.isfile(data_source),
+                use_audio_in_video=use_audio_in_video,
+            )
         else:
             raise ValueError(f"Unsupported data type: {data_type}")
 
diff --git a/angelslim/data/omni_dataset.py b/angelslim/data/omni_dataset.py
new file mode 100644
index 00000000..6f2f6fd7
--- /dev/null
+++ b/angelslim/data/omni_dataset.py
@@ -0,0 +1,127 @@
+# Copyright 2025 Tencent 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 json
+import os
+from pathlib import Path
+from typing import Dict, List, Union
+
+from qwen_omni_utils import process_mm_info
+from transformers import ProcessorMixin
+
+from .base_dataset import BaseDataset
+
+
+class OmniDataset(BaseDataset):
+    """Dataset for multimodal (text + image) data"""
+
+    def __init__(
+        self,
+        processor: ProcessorMixin,
+        device: str = "cpu",
+        max_length: int = 4096,
+        num_samples: int = -1,
+        data_source: Union[str, Dict] = None,
+        is_hf_dataset: bool = False,
+        use_audio_in_video: bool = False,
+    ):
+        super().__init__(processor, device, max_length)
+        self.is_hf_dataset = is_hf_dataset
+        self.use_audio_in_video = use_audio_in_video
+
+        self._load_file_based_dataset(data_source, num_samples)
+
+    def _load_file_based_dataset(self, data_path: str, num_samples: int):
+        """Load dataset from local file system"""
+        path_obj = Path(data_path)
+        data_dir = path_obj.parent
+
+        line_count = 0
+        with open(data_path, "r") as f:
+            for line in f:
+                if num_samples > 0 and line_count >= num_samples:
+                    break
+                data = json.loads(line.strip())
+                video_path = None
+                audio_path = None
+                image_path = None
+
+                if "video_path" in data:
+                    video_path = os.path.normpath(
+                        os.path.join(data_dir, data["video_path"])
+                    )
+                if "audio_path" in data:
+                    audio_path = os.path.normpath(
+                        os.path.join(data_dir, data["audio_path"])
+                    )
+                if "image_path" in data:
+                    image_path = os.path.normpath(
+                        os.path.join(data_dir, data["image_path"])
+                    )
+
+                ms = data.get("messages")
+
+                conversation = []
+                for m in ms:
+                    if m["role"] == "system":
+                        conversation.append(
+                            {
+                                "role": "system",
+                                "content": [{"type": "text", "text": m["content"]}],
+                            }
+                        )
+                    elif m["role"] == "user":
+                        content = []
+                        text_content = m["content"]
+                        text_content = (
+                            text_content.replace("<video>", "")
+                            .replace("<audio>", "")
+                            .replace("<image>", "")
+                        )
+                        content.append({"type": "text", "text": text_content})
+                        if video_path:
+                            content.append({"type": "video", "video": video_path})
+                        if audio_path:
+                            content.append({"type": "audio", "audio": audio_path})
+                        if image_path:
+                            content.append({"type": "image", "image": image_path})
+                        conversation.append(
+                            {
+                                "role": "user",
+                                "content": content,
+                            }
+                        )
+                self._process_and_append(conversation)
+                line_count += 1
+
+    def _process_and_append(self, messages: List[Dict]):
+        """Process messages and append to dataset"""
+        text = self.processor.apply_chat_template(
+            messages, tokenize=False, add_generation_prompt=True
+        )
+        audios, images, videos = process_mm_info(
+            messages, use_audio_in_video=self.use_audio_in_video
+        )
+
+        # Process inputs
+        inputs = self.processor(
+            text=text,
+            images=images,
+            audios=audios,
+            videos=videos,
+            padding=True,
+            return_tensors="pt",
+            use_audio_in_video=self.use_audio_in_video,
+        )
+        self.data.append(inputs)
diff --git a/angelslim/engine.py b/angelslim/engine.py
index 9aa07694..28fb0e88 100644
--- a/angelslim/engine.py
+++ b/angelslim/engine.py
@@ -73,6 +73,7 @@ def prepare_model(
         cache_dir=None,
         deploy_backend="vllm",
         using_multi_nodes=False,
+        use_audio_in_video=False,
     ) -> Any:
         """Load pretrained model and tokenizer
         Args:
@@ -116,6 +117,16 @@ def prepare_model(
                     using_multi_nodes=using_multi_nodes,
                 )
                 self.model_path = model_path
+        elif self.series in ["Omni"]:
+            if not model:
+                self.slim_model.from_pretrained(
+                    model_path,
+                    torch_dtype=torch_dtype,
+                    device_map=device_map,
+                    trust_remote_code=trust_remote_code,
+                    use_audio_in_video=use_audio_in_video,
+                )
+                self.model_path = model_path
         else:
             raise ValueError(f"Unsupported series: {self.series}")
 
@@ -131,6 +142,7 @@ def prepare_data(
         num_samples=128,
         shuffle=True,
         inference_settings=None,
+        use_audio_in_video=False,
         model_name=None,
     ) -> Optional[Any]:
         """Prepare compression dataset"""
@@ -145,7 +157,7 @@ def prepare_data(
             data_type=data_type,
             processor=(
                 self.slim_model.processor
-                if self.series == "VLM"
+                if self.series == "VLM" or self.series == "Omni"
                 else self.slim_model.tokenizer
             ),
             device=self.slim_model.model.device,
@@ -155,6 +167,7 @@ def prepare_data(
             num_samples=num_samples,
             data_source=data_path,
             inference_settings=inference_settings,
+            use_audio_in_video=use_audio_in_video,
             model_name=model_name,
         )
         self.max_seq_length = max_length
@@ -187,7 +200,7 @@ def prepare_compressor(
                     f"Compression method '{method_name}' not registered. "
                     f"Available methods: {CompressorFactory.get_available_compressor()}"
                 )
-        if self.series in ["LLM", "VLM"]:
+        if self.series in ["LLM", "VLM", "Omni"]:
             global_config.update(self.model_path, self.max_seq_length)
 
         if default_method:
diff --git a/angelslim/models/__init__.py b/angelslim/models/__init__.py
index 7202adbf..4036371a 100644
--- a/angelslim/models/__init__.py
+++ b/angelslim/models/__init__.py
@@ -15,4 +15,5 @@
 from .diffusion import *  # noqa: F401 F403
 from .llm import *  # noqa: F401 F403
 from .model_factory import SlimModelFactory  # noqa: F401
+from .omni import *  # noqa: F401 F403
 from .vlm import *  # noqa: F401 F403
diff --git a/angelslim/models/model_factory.py b/angelslim/models/model_factory.py
index 3427e097..8e029961 100644
--- a/angelslim/models/model_factory.py
+++ b/angelslim/models/model_factory.py
@@ -22,7 +22,7 @@ class SlimModelFactory:
     registry: Dict[str, Type] = {}
     series_registry: Dict[str, str] = {}
 
-    ALLOWED_SERIES = ("LLM", "VLM", "Diffusion")
+    ALLOWED_SERIES = ("LLM", "VLM", "Diffusion", "Omni")
 
     @classmethod
     def register(cls, model_class: Type) -> Type:
@@ -39,6 +39,8 @@ def register(cls, model_class: Type) -> Type:
             series = "VLM"
         elif "diffusion" in module_path:
             series = "Diffusion"
+        elif "omni" in module_path:
+            series = "Omni"
         else:
             raise ValueError(
                 f"model_class '{class_name}' is not in a valid series: {cls.ALLOWED_SERIES}"  # noqa: E501
diff --git a/angelslim/models/omni/__init__.py b/angelslim/models/omni/__init__.py
new file mode 100644
index 00000000..2dfd628f
--- /dev/null
+++ b/angelslim/models/omni/__init__.py
@@ -0,0 +1,16 @@
+# Copyright 2025 Tencent 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 .qwen_omni import Qwen_Omni  # noqa: F401
diff --git a/angelslim/models/omni/qwen_omni.py b/angelslim/models/omni/qwen_omni.py
new file mode 100644
index 00000000..3dce11bd
--- /dev/null
+++ b/angelslim/models/omni/qwen_omni.py
@@ -0,0 +1,149 @@
+# Copyright 2025 Tencent 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 torch
+from tqdm import tqdm
+from transformers import (
+    AutoProcessor,
+    AutoTokenizer,
+    Qwen3OmniMoeForConditionalGeneration,
+)
+
+from ...compressor.quant.core import PTQVLMSaveVllmHF
+from ...utils import find_layers, print_info
+from ..base_model import BaseLLMModel
+from ..model_factory import SlimModelFactory
+
+
+@SlimModelFactory.register
+class Qwen_Omni(BaseLLMModel):
+    def __init__(
+        self,
+        model=None,
+        deploy_backend="vllm",
+    ):
+        super().__init__(
+            model=model,
+            deploy_backend=deploy_backend,
+        )
+        self.modal_type = "Omni"
+        self.block_name = ["thinker.model.layers", "talker.model.layers"]
+
+    def from_pretrained(
+        self,
+        model_path,
+        torch_dtype="auto",
+        device_map="auto",
+        trust_remote_code=True,
+        use_audio_in_video=False,
+    ):
+        self.use_audio_in_video = use_audio_in_video
+        self.model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
+            model_path,
+            torch_dtype=torch_dtype,
+            device_map=device_map,
+            attn_implementation="flash_attention_2",
+        )
+
+        # Load tokenizer
+        self.tokenizer = AutoTokenizer.from_pretrained(
+            model_path, trust_remote_code=trust_remote_code
+        )
+
+        # Load processor
+        self.processor = AutoProcessor.from_pretrained(
+            model_path, trust_remote_code=trust_remote_code
+        )
+
+    def get_observer_layers(self):
+        names = [
+            "k_proj",
+            "v_proj",
+            "q_proj",
+            "o_proj",
+            "up_proj",
+            "gate_proj",
+            "down_proj",
+        ]
+
+        observer_layers_dict = {}
+        layers_dict = find_layers(self.model, layers=self.observer_layer_classes)
+
+        ignore_layers = self.skip_layer_names()
+        for name, module in layers_dict.items():
+            block_condition = any(name.startswith(block) for block in self.block_name)
+            if block_condition and name.split(".")[-1] in names:
+                observer_layers_dict[name] = module
+            else:
+                ignore_layers.append(name)
+        self.quant_config.quant_algo_info["ignore_layers"] = ignore_layers
+
+        if self.quant_config.custom_observe_layers_names != "default":
+            for custom_observe_name in self.quant_config.custom_observe_layers_names:
+                for default_name in observer_layers_dict.keys():
+                    if custom_observe_name not in default_name:
+                        observer_layers_dict.pop(default_name)
+        return observer_layers_dict
+
+    def get_kvcache_observer_layers_names(self, observe_names):
+        names = ["self_attn.k_proj", "self_attn.v_proj"]
+        return [
+            k
+            for k in observe_names
+            if any(k.startswith(block) for block in self.block_name)
+            and k.split(".")[-2] + "." + k.split(".")[-1] in names
+        ]
+
+    def model_forward(self, dataloader, **kwargs):
+        self.model.use_cache = False
+
+        calibrated_cnt = 0
+        if (
+            "gptq" in self.quant_config.quant_algo
+            or "awq" in self.quant_config.quant_algo
+            or "gptaq" in self.quant_config.quant_algo
+        ):
+            device = "cuda:0"
+        else:
+            device = self.model.device
+        print_info(f"device is {device}")
+        if dataloader is not None:
+            with torch.no_grad():
+                for batch in tqdm(
+                    dataloader, desc="calibrating...", total=len(dataloader)
+                ):
+                    inputs = {k: v.to(device) for k, v in batch.items()}
+                    try:
+                        text_ids, audio = self.model.generate(
+                            **inputs, use_audio_in_video=self.use_audio_in_video
+                        )
+                        calibrated_cnt += 1
+                    except ValueError:
+                        calibrated_cnt += 1
+                        pass
+
+    def get_quant_module(self):
+        """
+        Returns the module that will be quantized.
+        This is typically the main transformer module of the model.
+        """
+        return self.model.thinker.model.layers
+
+    def get_save_func(self):
+        if self.deploy_backend in ["vllm", "huggingface"]:
+            return PTQVLMSaveVllmHF
+        else:
+            raise NotImplementedError(
+                f"deploy_backend {self.deploy_backend} is not supported for saving."
+            )
diff --git a/angelslim/utils/config_parser.py b/angelslim/utils/config_parser.py
index 6b24dd10..59e3edfa 100644
--- a/angelslim/utils/config_parser.py
+++ b/angelslim/utils/config_parser.py
@@ -93,6 +93,12 @@ def set_model_hidden_size(self, model_path) -> int:
         json_data = get_hf_config(model_path)
         if json_data["model_type"] in ["qwen3_vl"]:
             self.hidden_size = json_data["text_config"]["hidden_size"]
+        elif (
+            json_data["architectures"][0]
+            if isinstance(json_data["architectures"], list)
+            else json_data["architectures"]
+        ) == "Qwen3OmniMoeForConditionalGeneration":
+            self.hidden_size = json_data["thinker_config"]["text_config"]["hidden_size"]
         else:
             self.hidden_size = json_data["hidden_size"]
 
@@ -147,6 +153,7 @@ class DatasetConfig:
     num_samples: int = field(default=256)
     batch_size: int = field(default=1)
     shuffle: bool = field(default=False)
+    use_audio_in_video: bool = field(default=False)
     inference_settings: Optional[Dict[str, Any]] = field(default=None)
 
 
diff --git a/configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml b/configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml
new file mode 100644
index 00000000..22e99a89
--- /dev/null
+++ b/configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml
@@ -0,0 +1,23 @@
+# Global configuration of pipeline
+global:
+  save_path: ./output
+
+# Simplified Configuration for LLM compression
+model:
+  name: Qwen_Omni
+  model_path: Qwen/Qwen3-Omni-30B-A3B-Instruct
+  trust_remote_code: true
+  low_cpu_mem_usage: true
+  use_cache: false
+  torch_dtype: auto
+  device_map: auto
+
+# Compression configuration
+compression:
+  name: PTQ
+  quantization:
+    name: fp8_dynamic     # Supported: fp8_static, fp8_dynamic, int4_awq, int4_gptq
+    bits: 8                # Quantization bits (4/8)
+    quant_method:
+      weight: "per-tensor"
+      activation: "per-tensor"
diff --git a/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml b/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
new file mode 100644
index 00000000..e6f3106a
--- /dev/null
+++ b/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
@@ -0,0 +1,32 @@
+# Global configuration of pipeline
+global:
+  save_path: ./output
+
+# Simplified Configuration for LLM compression
+model:
+  name: Qwen_Omni
+  model_path: Qwen/Qwen3-Omni-30B-A3B-Instruct
+  trust_remote_code: true
+  low_cpu_mem_usage: true
+  use_cache: false
+  torch_dtype: auto
+  device_map: auto
+
+# Compression configuration
+compression:
+  name: PTQ
+  quantization:
+    name: fp8_static
+    bits: 8
+    quant_method:
+      weight: "per-tensor"
+      activation: "per-tensor"
+
+# Dataset for calibration
+dataset:
+  name: OmniDataset
+  data_path: your/data/path
+  max_seq_length: 8192
+  num_samples: 256
+  batch_size: 1
+  use_audio_in_video: false
diff --git a/dataset/omni_fake_data/fake_data.json b/dataset/omni_fake_data/fake_data.json
new file mode 100755
index 00000000..885bced1
--- /dev/null
+++ b/dataset/omni_fake_data/fake_data.json
@@ -0,0 +1,3 @@
+{"messages": [{"role": "user", "content": "What happens after the text disappears from the screen?"}], "video_path": "./videos/0.mp4"}
+{"messages": [{"role": "user", "content": "How many food item is shown in the bar graph?"}], "image_path": "./images/0.png"}
+{"messages": [{"role": "user", "content": "Why is the speech described as rich in frequency content?"}], "audio_path": "./audios/0.png"}
\ No newline at end of file
diff --git a/dataset/omni_fake_data/images/0.png b/dataset/omni_fake_data/images/0.png
new file mode 100755
index 0000000000000000000000000000000000000000..b433d7dd239baff730b73b8b13b86994d835b2ca
GIT binary patch
literal 44922
zcmb@u2{@GP`#(HUX%UG^){;_)P{>*lS`gW?MnZPk_e$B9BqSlE?2>&QLXwaeJ7dW<
zV~l-_F}&wp&-eNLfB)lmyzl#ekJsVpDQ51uuKPN#^Rt}ib%)$mQ$DbdZ66ASI&fV@
zQ4@uth5u;u_UwkASbguspil)@t}9;A@_0K(^><;P_T}3^Uk`bCCh^0r^Ea*~I*q?j
zOw+cNn0>BbJ0Yd>wr%oh-e<Mf4ugVwr$=_)_>AonFvBlj8|+hKAIcGW7SR=>7~GMy
z!?;Y=j1W=Q7%!377imqT*<1V2;q6T3?)vu+4({En@Gkn)KGEy<YW;t9SnZZi8H+mh
z*azLdSh!%bkT4dIK>o2Tlk)Hm%e5U$f4;8Ulu4m@kgqhI2F1`|<m=n<)0&EZzIKa5
z|M?w{XmAM2pVvoUx%K|fi-PIRc&`0<k@-R9vwvR|@LJ{XRgX3PpIvIe&K33dS~aDE
z^5(94K5UFS)Ut2;ihi2!C{q5<u+~(fvxJPRgO#~`S;~eU?dS6R&E(?drhiB|^>$rL
z44>23mt8Vj>k~5s!te_5bz))wme?yhz36dXMy7fe`{3Czyz%^aQ`q_-nrcwJJ|2e3
z9cHM%cd*4=lQh)b{pwWVoTBKU|4B-SMSt_blF-ghjp2#Q#~*KP`6;A0NhK|ieG<9Y
zHHroud9=6gA6idOv-VpE;2!;zakJcI_K=Wa`A<j78g2Oh<K^j&91*)~qMcZ5l+}e$
z?(OwvkMR&b$Jt*Qs@svP=kE`fC|)I#JN&sQB%-q55e42ug@VybWS?@6MV@5;Ezd?X
zk(*|HE#{@$N^8t*pQHK($v#|#dh`X=OZ!9<=ux8!V-4k`5pO4A4N0G!Lp#$|O^Yzz
z+1R*qzIs_LJR;(v;a8pHJF2RxV?!0r>tBx6-kT9~c6RpInEi;E*xRXAJeKO)X{AM)
zn8)K4jvu^t=$iJhYp>DF{QTFkhM;q@^%awG29E0srY`R86E#F%Vw|tKr6pe^$EByF
z>S<YnHBP03lKC&awQCnS<a}6_BRT4o<u^1|R#(ShXa-v~t1D}3<FwrCk26H(%p?At
z2R$-~-_gnoF)5onQ?^z<id?+-yz8U-#M+=kvo3l=duPCv9EJCyomMRNWd^=;ZdFVX
zN&+u4!;T8`#I%e~#+&5peeLdMk=rJBq*qn3qZSf6e0HZ$=Pb)E%%4<_tebUV+qmWL
zFCRnZ{9QI*ON(|TahBs31|_O<6CEm<kbi2Y_~Y{&@;$cl8ypHYe9n<-NemuxkNn$u
z=Y1U}-3xjHz28eoOOLro!FTkn(@rx#Q{&`4T3#n4T;V<RFf>&^?|oU>QFS@;dyMJ3
zq9W!f*|paXv81@N>ZW3%b>k0LdG~n^(S&d>v(EnhkUJmM5|)M|UW>esw$3S|;t70g
z1n<NL+(jB@d*(B4#yc+kzTRgmj(Xf?(a_h&dgREF7q?Es#~w&Qb@j~7%{BJ*y&eeg
zXgDCD#c2%V+K)oQtvyLP2<AIuEdS{D9{Y;PBR(?<+}1er$sPlT5~-a7X30@QQQJ$(
z#)g$XhqpJ2QYK)w&GHZYBI&k|KjVJ8y;vFNxvVNNI%l$d%x{=_cUZJ`rP+9h+&;L7
zpYO09@jkhg(x*G(F@IFE6-}BiF+aQXqb#R(H_8%CX>e^aj%+#|Y}LBC@=<;lYGq|b
zk7KWf4Y@PQ@7c|Gq54^u?CY=MW-!zs|8=?|g_C=Q;`gHtMm6Wu`mN2~UBzZ;Wsk33
zO+_}!eJd-go8u>6I?V+L$*fv3dn_<s^}tiUIueOnJM3$hcWe%m{GJgvG1nDp6!l>y
zCZn<bn|mXN{cC)&K3x&(NtB?4(pr<QbEH1Hp!5Ewro-0<b?z7Lg{%}~RPDhezh2L{
z5aHlM8zr-((xeiZqgz0e>Gv|P#D9A$bow+OAHID%rzcVO49h|i(T5JwLDKTC-rgW-
zVl}gUS(UXYpN+jeLx=pf%+?lh+c-%8=OX3QAxr8esNOss<1w2N-#yMdJUr~t;XM+-
z=rc;|s$Bb{t?dL?)X5`Z7(w2mWIOX`4Pz*jx#+Rq%bvq)t<%fpj$<@(YlTr`KN7@(
zn<qR*pU}CaY%fOnhu7BDa`rkWcT9^&NCb&lwvWzqreECNou(Q>qlA12mNKGSmZxJA
zL>hKCX;{-zWBCz%_IoFei#kM{SmaCo>Wuwma?oR4T@4nzY+{T~K3rIK^+341qdb@Y
z_$)rfduAe7uBwOKrFX1sLbt%EULpDZdXzugSNUz*Jl)wjcmH5)h5w$Yn#GFy+#Fx9
zzC`79jE(N7=cwN**_&8P3+w-V0FA8@uU^P2t3I|(AU|V^pFT;-X|2L1IW%wX^6kK^
z(~>4es)%)rTTaU{BCX6>OPm>fPW~(D+!)uOurZFS#}C$RQ7Nvmb@Nq3x$vkcU4l|d
zLPGt1;p%s4J?HqOql;uBRe00ngiMskj5P%XjLK15O%&|5-^H&>`*a{XV9D+7e^xbK
z$mnUubU@c^x2oJ>Y`W3M+FCi8_0a%H*cbbr(`-#SZ0#0CZ;A>TRX%G7Vw|{>vZ2+$
z`N%W&(#Nh4ETsn9Wp^L`cE34t){P6ZB@T(!vi+?_kZVLlg!Zj>d2WxHF~g0K`2*t2
zFI#vmEG)VMwjf1{$jA&izP?Eo@13WY7-(;730|)XPVzn((UTz~G`jI*tKTxENuJsz
zpViSTeJQJKezG8B@3#(_W%~9M`KbAMdt|?6OOCesM6J1}$zI8FTXKVRE1A?DNO!wa
zsj&s)u)XHcYdbNCzE<elJAtI}DG78|zDb^&Zl22HuGz*Qxs}vVQ<+36CGKG}WXs_W
zHnKGPUE%s~YXv1KFZxN}=kA-YwW9b-olQ)*4TX+w&8BQURB(QrqkS%SDpI;9FvYOa
zWp7MO%;Dz7rlxDCUT?W#NI&#Vy!@QT@S7C!56f=rpceB(kn4Dj{m7H9y6d-2$L!f1
zu;3ws^Si@x7XFG`ymFJ?M#g_jX2@PbEk*7?V%rD5b|NXKSt(EdIajtwWkmR|ts{Qx
z4cxfEY3f3dd?RIqBsb(aorH5*0zj@LP#Cut7`3-c9~V+T%v}l$Uz+IlB@LH7Z0O11
zoWM0P(eHM``jTb`z5>kKlxi<R&Yu#qmS*Xw0T}-^I0pARc7tkv)Sc5$dTSyjFsIgS
ztxCSJhFC-EP_x?RF_#*JKa|zXwIR7R?`@+nz17O?DdInN^svG#Plq2nqpSx#^1Aiu
z6e>;gy!%*%L-S%zZ`^{(*0-F~nrP?kgIt>d2j_i>oE}1(XSlXvx;!Si{huNioUfrU
z$ZYa`9E1H4zFkFG{4T$>v!Hr|d1%NrKq&4$w@X&5_vV;?lrXhPLfnr1u2iwZqeq8;
zchE6%DPCY}0x;w9S-4WYxzxHoG%XhdNS96I<l<5~?nn*1D^Ka~SXeFa=XCIFH5gs=
z_z8*WsEH4g3HiG&e=Q#?J<p*KBt_uNhav7!HizGT`0(L9V|YVD!|OM1o;FYH_~fQ3
zdGVrlVYs>IUNknUta|bs4pt=m_@R&)vlX%r?lWpsz=XVSHys@&h8S<`wYP4domfkm
z@3O0cpx}`8Vu4h{4XmSf;kocA5{KV)YoD^)m+#=-S_pE2pnWUv&lM%NVb$QutUc98
zdarp_W_xp)m7BX6CP(uGrJ2RD>4&yg*4!%d1Ono;iqrO)yU$IFTFo}Oz=p}QsW2A}
z3F7u+l?POJ0yw+{yyZSOg_S3IPg-Z@d<}%C$5IO)2CWP?j8KM+ymvl0At9j>Kf6p`
z02)<`vtfT|c~scAaaxXaNs)&sJUslKwe>!et)+O^!FOy4uE)5zzQ)(CoN`?%6hGz(
z>DF-~LSp=TtbnB3oZfK2l^R!JhcF0q?B<jNpgsmYnILA#W^YoMtB+>3JLC$3g3Ath
z6KimlSokwok;h}Mu=(~G7MIq;XaPHR&&|c=#pRT34o3M+M-1)I@89n?5`fs;hhZT6
zulHh<1+2RXa4{E(;9}V?222HzJ%q#}OJB*e1LC3aXzjwTTWV^!xT=0MQHf9v3<p@>
zpT*v4hfMbcM&*Jp8d=rv_-}hW&Ca;SOWV#8VN|YrSG+>7O6e~t;)$w81$u&k@*C;$
zS49&PfhRYG9c2n@8eJx58#AFFP41(4@_0jba{Hlr>Hg<5YDrQVr!|097QFd%DM?Lc
z<);EdHMo3NIzzGdPm+C%&jmx?%M~g^qq(`s-;}#uySVUAO3)MJ<!NMxn%dglBACt)
ziOtA50ChacN^fRwuSU95=r>Sm5*BS&Y&<SOr4s#3NSpmdmW%W5Xm%=<ij$L#R+$-D
z{w`23+AdAZ)%(ziBmyPb=T~<Cxng}<f7~8Tbz3c4j&!e9kRm@++^qfh;RBAV0ii(<
z3=sPOBKUzotDKK&Nx$p@kX@8=)&V=-ciLIzAmw|~_*A0(&?{1LNy*p<mP#wL*4Sn3
zz=GNw-Ha0PH{F47KvzAiGgcq}K<cOO$$6<ww;{@i^v|B2!0Fn<&G#97IT*KF9x5i2
z#gi#I9L?2H6RC7jSt377)^{NZ;qdie3XEJ04gQbVI(!8p;XT-H-hH3WM@xCH37QqM
zzE$J<`t@sfpzrT*Og>97DO&#Y`pC=~=P<e^P4}Z@1INZp=@}XCo0-wn=__<hFXJo`
zwF5f?z{0iJs;Ix4JL-p8;~A#~$ExAVfaV&~NEiUx$%I?uuqknr66rYv&J=+Bjx0vH
zqsrwNYKw~Q<UNYLb$am$qYvu-{rjx24HhSwF(lc=M&|W7OZhN5E)NtW3xt#Z93GAU
z@}|w<lHKXoajtAAjWW1GB26qdbJKl_duv@*wB~T~`_hOPJ`Z$)P6}~uF4xoX&|oe^
zDR8UcN@o^bqbAJ#!Yzo|l+E9KuB}@$%JTc}5pKAq)B4ryyM}*gi9)iOk-I3*I2${A
z4%tYlCj+=N^7f5tNiw2;{I-@;bUC@qU4~2iQ-G{}nO-*a@L*+>SwQ)1ll-|+3t#pN
zU)W?gHg4WBu{cxO3pZka*<b{4$q5I5`)*QLfT<_@EP>F3L?J4cumtxfxXEk=kMHvR
zfJU1P$9-<ZX8WX^%C`-RIYkJ`T65qD-3i*|#F4OBwQptPB%=aOd5C_T`#B*Y+*bSe
z`<%_+4zl=?WP*u<d}r_vIVB79<^=aP>vTs-GxDK2Z!;5Gfc0VqAyGoO%8@EM#>RrU
z^SyUbH_6M|L7UpkAfwa?cXM^RN54g%guyfb`F}5zX7!Ny{gV0h!TZYr+?$t4a)syu
zo&HXXqO9w6UFz~ii<N>W2d!!voeX*F>Q+sC&8SV4R@0gYvn{Au)5lONI;p3RyGLq7
z%{)I8lDAfNhxnyQ$73qN5|z8!7#FpV$?hU3Cs1e68`V}O7$x422LSt7^PUZm3>JUK
zsAo@KmaZ=_uKIkINCpI~p0YtA_&%x1aV|h%-2tnX+3G!S;N-zo?w=Qrix~4pL{z2X
ztxXCrOgxuO&1#Rt1hTz~zZ}Bo#i^*OtJ9)@n=ooupJrf4xTO^DXcbGmS2A2az<hSm
zLB#m(LAqw8iH=@{6}f84FfreQU_|sf`WPRx&>FJpD3x@vAoNmFd#JrlVVm6M7$e?I
zC@CrFFkT4%m0h6V0tm}&=@FvOw6wN7U-o)1?yhK!uUR4@QoEXj0}oiKv$nszL_4}s
z$^vQ$(}xK21?dOhO2tGP)2gbfa4jl%0kNH~vqYk<FpUotZ=x)h(riM-n2Z58?EZlD
zDE5(U#Fq3d=$%|5S9^V4*&?CQ<E|p_PaO%%qAQzD@#m@=kW2S5vJ+1nP%0)YWqnMh
zK>_4w`Fx$}b6YH{oKE(3=|B_g9B3V>4x8!R)IABeK3I?m>B><wm)l}~<*ob1m)$2^
zw<XJ2I68j9iS`s3Cst4;zJK$(`$OFe*3Jd(?388^2Wf0P5xaSMWI36X84~k)urrfn
z<i<LZQ&77P<?6jdu_M@RPht4Th%qGFbfC#}3>-geJ`Cp%jG3mr;k;cH5zKopog_Sf
zBP*MEoQ7RY+Z(jHH%&Hl%(repJX4(4N0!p|NOYQY5_@*d7fl^R8!dX}QJZv_{Fn8c
zr8{*>uIUw*DQ14`5pBBmUfpjgM%unzf2H4H;zUPb=(I!atP&zEcsg`+n#9KENIg>*
zLri_YIAhgD-RHT#I1E#T-um7T9^HJt%ZTyyI;p_eAiKgR+OTEiZC?B-=l6|NX4E}w
zt*<ur<oB#?5R-YA=8XfbRwnnx^o7J@u^&eyjrv2L7gT>1Kbt?&MzB7mnKz0=3$SR1
z^>_MFeHN3LVX!6A*lA)qrZJ^ra&pGbo(OHaC9A>AcT(wP&uV@vEu+k1Zt9G(B(PKj
zauH}dSLVDu##ms&neDdlE5D#TE4v(fBQGFnM*AQ^v&vUOKTWqPcPZotB%a6ia?z5H
zb*m2UFqB`kDoNC1Gb(5s+Dfed=0`>bQ(0Bzw6*D3qP#HTO>7!=$J*(W{hR{2Df3l~
z%~6m&-3+TCvu)uDw)O$Zh*sGd78X`{NlEl1V{yLWL!WDR=i*ChB1D^SB(~`|f9rNf
zQ;bkH*Jwtz*6^eZrt7w{1K81_p|@kcHJt#Vfa`=4WJFqXTza(AvLjhL=pv0xvt(z^
zTNU1gI$Fk2L@8mye&G8Z>|ePVlzeegD%Q5E%6GrHQv`R-@LlfI@0OVI<#uWJ-*&qJ
zz0PluC(qP6Z7h3iw)%%4=79c_;2^n3@_zKs*rRAtjeF+u2Xk3co}`eG@t;NY*y=hv
zyd1qP^Q3HK!$O9XS7KTaZCU$VS)N3ZqzrTCA1?4%U<I}<Z>!hlu=RyThP#5>WFyRB
zG_S1o7%*ymVy{IZZ}hMEGi4j*y5VAsy8~XoeT#S<Ue7aNmyYkdsjjZx#vL|y9_6(3
z$w;8m)~$o?IJPylCbZuw@<~FEn3Ly8{8E;>?>?Yh=Rw}_x{i0+D*jMnxHW#GdhYPg
zrPO$KQ}esyDEn2_Ac@4z)qx)b^s(@{cxgxr6&k{|>pX}SLvSO<gqr=3#K2PHi6$qa
zv%G5oljKs=^Q4Z3HSzPu0vR&e!L(?Y&hM#Y`&;kq%Q_$&w{G3CZ4t>}5-Vl;)N|8i
z>c=sJ$N@H&`;Z?sHI0!VBLV7wya&`(;*ZUidU+o<Tg0IdrJ{q^KX#J6CPAOp!pi`5
zFjQ4_HMQ}t><+nSf<NEc#FYpjCdguj{MtC;&{S60*TEhF`VpsJy#obu<c^0>wYaVX
zm@)@W@goTv?5#qx*Z0WFQL>8{>|tYWF1Kp9xDP+xUL!mcB4<y&COOF*^<&wdjH(Wn
zvk??O_TjW9s9l64FT!oS;cG9e8x#HroK6oJ;J3JiNQmD8>sVnY=dp-`y6=9>9~Nm<
za=(WOw(ALAGf%Dl9v#cTC1bTN#(p;*vZC?1BnrUEI5<#tZ|qJeytO)(P%g{>4q1^V
zIw#_l&B(XUicxLA70WPfl@a;97WUh1_gv7nMg5MO>rUa3kqB4J6JifhTmUcafx^jC
z<`>vXM8w5EaCH<<miVs~rJMk^W9Q(_j^OKT_ON!&OWm9xjhk<8*~HAOatKUMO)U&L
zcbrIU>v=nI%DG-M@jdoY>$yNPRJcQW#KCkH#0Ah3)G0ZHt-@06!Da<h=Esj8TH&vl
zdWx_^nHE)aqe1eUra<<6mJ_j!KY!kCaw_Pw;<j@gM?^cwmIfpDVeQo#*vE()yS1Xe
z?Q+M=K^yHw_^$s&lkZ`O%rYTc0=?_8H=07{8gC+SEPHJe;f25-XTKBIji~Mj`>bQO
z+1{!rb5G<E2P+l(x-V%0Ymyl8IO*l>jd7YP$=46%yq34yUQslm5wS&d-Xy$v^99IT
z*<9nD@J_qQG4Ivj>u=GE?C8RWtsJbZsGsY&cvk)$xt#Qg&smF7+AK2Ui$+{80spiw
zAKP7Wz7F^X!oGZO9vaF^OOn2A$9+6727n`nug0wXa-KL(K7W{7)^II5fwQpR*4i~s
zsr8UEmvzZ55jsVKjDo|g|7N{Qtn}H-&P|)rDIGzG7XQ_HK1Q6}hxV8M<7z#j<H|fU
ze<J&$^|X$*K=5rf`Y^$i*0TaB9TO|FT^{ol(~fg}B5$Q!vn>vNAUNe;RMw}@kn>qz
zsPrldw@;7wd3vZ&FBf;ft!Y8-o}G|IiKkE9##$nKhLM`sW+Zldi!=c)v-PQ#BFhel
z83ss!e<(D1n}VjhO`*Bs?~nH4@&M7cxfG9XY+#fRw@mTXA<1bNe#z##W6k;DlxCV?
z&UsN$ns;6pnP^4rM3)+_ap}@=#iD`hNp4>(Ju`Uj*;%@8xI9oAw$;qiY7$ROOuQ4u
zKLox+;+m0}s861q8(5hoM#0w!;1^|WNF*gBWa(FW3WQubT(a+dx3#xY?<t5@NX4~M
ze3l*_dgq?h+@<pq1Fv`WnZb|PpN9_TBJ?JwBSd^@sa06lHpfS<KxewUwAJf5XZSp$
zqF~9E4oF{w6pyUtyJyvKW2Vf4vx6}FFE19Y<|6y&kx~%Hv0-8P<sV{)MM5i#2O1y1
z9^#a$O#n0IG5(s)cZ`&pR8o6335B1V(-Ky)^1d)%OR-<&-SWm#j4?zeQe%$J=(}jp
z8N&YL*7Z6%x!XmCRXPq11#ir$QQM31{KiK~BDC)kZOor_+VLK$2T@xfkZ_Pv=_S#J
zITD{4`zhZNU7=C3kG!Wh-zaCJBEr5%ADGwTx06wI6N{rC679br&i-2&&qMOmS@l1;
zhLD;Y8*bfiD-N;^@$nJMZF>rldvY5yk%B^3zS`a3xhNs=@rd&`i=xBwZ=9#w$M7cG
zmGhT0vc(c^nLT=>DSI!39km7(2j-OZXY*npPpQcsGR96Rs4E7mbb3|Ru`TA^rflPA
z5Qxa2ZN{@CUwkzak)g~LftRmAjEDM6LEKJRSKO=lMk!b;8xzVKd;7SxRj<8|%lTj5
z$Z!!kx^LI&zFEjw0@+7G1-^j|T%8&d4>D;CAQZRMcz_7tzqmdTh0p)hO&PR$IMI`D
zgM{C>Bd+`GsxM4QOiXMYC}aVbVPl)R82Bhlu?3m_-issVqW$Mvy6oeZJbJs@kvj&L
zL2ib+%;GcqAQ>^S^bzlIP!E-pE%`S=yn|u}D@-8!Ye>vqCt`|{|AZIbxZs?%cbmDN
z*nr#JJj2cPbx8(pxp1&f{o?f;G%O6IpWXU#Agf0Ry*Vk2T$@pL<wd>`10!S3sV60L
zD-{)&hJ6F9jmSNRbgwpHZwTIedw$5V;Q$oTu0g%$^y$-xDvaS%&mO!{EgaJV3##?@
zw*4Bh%2*mM>ht=ARWDLNtA%j%D)Y~y?!4`fH6Ru#t+kGiH-cT~if(~5iO!WL4LMss
z^yxpLJ5Ca8oLCI~JrBJ23^G{ovoy*G`<_4bdMo3zeqIi^+YMhnN8i_Rr9VU5u&bB?
zKkYI`CTtgs1Yw|oK{lbjJ+LHWB#LV@c4@m9c}q54?$V{Ziw1N(p|o1;#t9wMt{IIT
z$+9*xinRmc8{jmAA;uY?zS*!hO%IP3)q42e(Q?0We=sgEgdHN4*qy)IMR0_hjaY>t
z<=caDjfS;a{27Do_+g9b_M|%`gGtKqD??MG-C1|QTl*5DE*DZ*C^9!#re5<?V9+}|
z&xc>A(xlvR2N*S1&$H$Z>Rd~-&j@>MFotj*#9$1RTQ+dVrcw%GOU{O-*_!WAn&)MB
zXQpW)nxNI3Oj|+}HOaa9v%6ahBi$61%KD|KLoUq0Z#fb7Z1s`Seg&3&yA;zkN?2yW
zf5ya^nf@(HRBt2$8$s;UO6l(2%2r=b8BA$WUmV7G|4y3MNlAGXZ8)a;!AWX-e{Qz1
z$CVl$GSH3!Gc)Hrv?>{^KSk6U;mWqtC3u&(-z{*pBt2Rk_@y-1kLbfPe@a=U+hTN-
z&rW3Xy0+%6NoSS`O*(~i*fe|hhF*6Y#^;&Ttm&$jlxH{T{4Js=pW_zpsp-KQ3Q|B(
zHe<J`*a;|;UL1Tt_KGi?i1J%WKa4rVP8H8?@`k%V26e*AO<M|vm;5JVb-T$zB_rT}
zIi%YSeZ%&D^V+La>Yt9iTjN{s!$Hde4f`U1N+C_)Fge;2JzR+|Ti(C<*20pOS<+E(
zB<J<P?LYQXg8nro#SFB*vri+owxFOOZQ~xOJonmV&c+F+RU&_>Y7JILc9owJE3i!n
zah3O8%iS=*L;2GiwALHDyb*Hn`e3dlJEF58C!GP}pyj@;e-kbO7`&Jv?3-oJ*vuzS
zzCJ9w%SPlg_SL24%rf*(ziI4d8x|>4eg(Mgb}p%}zcO5FY|EWKn5$tBefFEn#>zva
z`mEg{3%JteeJ<3Fmr{0^U|V&B#ipv}vzfW9<@QDgN~`@&wlcXzV$t2<)kQS<1H03-
zv_o_J);K#t()$-iw2%s45teF-p(y1w)A=tCuM=Y;9L9}W1DUfd_`Ea_(Rk0rMa%vC
zL0u=1#OVSeRb3C6&!*GS^Q0x6VWl6J@SBjx9#L)=9Ej<(=1WGbHekf)tR@!yHl@<e
zA3x~*Ho8)l4C_%vr3vNl5U*Eet?zziuj2R}{MBYCH$#<kh4?h&Rq{-P=*zw{^5hTm
z-U#>fnT@I`Urs8U=pA15?(f)IO&KfXOCBjsoU^|W{dq31+TxPu5--BW78e&o1+3G&
zSzlDe?`C4ss6a}+u{Z2+)t##c8-8W4T7h=bAk-Xe$aGIzwNHnt;Zb9c)X6Dv{gzz2
z(}6Fzp#J=)w1-cdQ7tjBd@Jp)MtCyZf)0!L$w2U4NsU6xiAKboO9vMZY6RW^>IF=(
zI~ZNu$Hvri+ZL-p;EZ+#$>kfgI}H2M&HGFfkg?||&|tmri2@^*AM8?8Fl^uXf~z7N
z%Dih2GIjPaE4)U8WNVIh$jjeZS~nzwr@&?MSm{*m93DQ9($VXZG22$0Y^QaG1u5`?
zjqYNwcG>oM-iSq7-A}l4TT8;*hq$8X;HK+gu<u?Ek~fz-@KZ9UW3J19Ifqo@r@;^B
zq*nC>BV{*IZFINH+D`~QkOaRbPZjT2Z#PN0aKRr};n3tXw{@Z#v0-ld`*18D^?PzN
zKE8XhdK>12%)NlyPkLl~M0%me02^N9PfbCJAFt61`#0qa3s{p2N;KkzUi97?H<sp8
zPi|C}TVis3WDS4=8}+2?mwB&jCOh4MmMO@B4TIP%w^sFo!NG4-;RU_ut<^Wq!nt55
zL(zKihgiqxev@^Y$0Y(w5mQX_yh>#n;MjvfXEXb$O{XY!F!%PE(P(9^de~)L;Fyd;
z`7hr)^4H*C7+?&~k+3mHuGWm*kl+?m8GZK<{kJLPsy@YNOLQJXmGs(`qlhs(YwB<2
z*Db$2ceEUj^;7cQ>Z2+kHU7muX7veV9VrY1*)$o(p(rT7jNSBKjA+wO3M}>ISp{wU
z4qOu9nw6K84LkGI^<xwID(+|r96Vl{@Iz4@w#9o;D`s*^VtwK4tq$%=rA*U`$I4*b
zQ07Mp{;3^(wxz0xN53tf@C3%cZ{BN~;4yEo_|@U-!1)@Jn$q`M6vX&2aPSmE;^<=h
zovqroH`1qU2Kh8mjQpP)BXEB@Xns6`;`=!AQvc1a^O3$kpCi>IF7NS&n`I7s3;6Bq
zjUm2%l3j+K{68HqZ@P#>@_!29e=6eN#IRo*Af}zMp%V}RNWIE&@LcS2GP3<Rs{Av@
zD5V3DGk;rn(x%>R{!YZ9O~}nDqsXh_PSK)jSGt|oe?gfBiF#1>n#4=HQ0g+P9EcTA
z?n365**+Wm&3`S6NC6k{buHK6mfYXopzhb56qORGk^eO3m?w%CcMwS@<6E<$l$Dh9
zx27zN3ywl`D<~*{1_l&r3Ss7Zp~8D!RyN}DJ?gPsrUBxJExbl9g^`w)Hc&oM?IUm-
zrae3Fl?_T<%#X--zQA@(rdDV8fx!6w=T|=s5}=+`;Gt?8?<h%sk5DEeD*7}vHB~jj
zR18!c<n-B6!&)Kshw@4Qk$r@<P~RJ8Sg!KN(x8xktS)%fR5ni%Ltre)$@yfm^>>o%
zIMLQ6p`<5E<A=+z&Rjrv@vPN_eJGsYah83ij*gC9QN9(141Q?;n`;m7CH>7Ac{1&o
zqPvvy3BIoy#xnKLN|dnGImkx`0|NukU${{G!0xl5Sl+f*FjkMe#qpCE({w5I2R=lR
zCPeYein!|3fFsW-HWSZ@C9j%qlaI93FBn9}ULVu(b1zAPc$<-Go><J5mJy?;r@ugO
z8@9E`{iqTOH`Aq~*-S)GNGMwbX7=priPd+pY9DjB%1ctBnkO#Go2Z^)dD<{`n`8t1
zJhJr{KmX9jcb~Daw%@GEOVMubYwHWbY{&*(JuxiD5BByNUBuc^e82LkLwoxOZmWHk
zN{O}~R!tb4xC|oJo>qF8sH;c4@#yvTwX)g+TLBjD9JcUsiJfG$lq*gPlS6oN8RduO
z7K*=V=Ll1aR#EhgTH}9F&bI4vjn~+s$BWj<?i?ogwd@m-jV&dta9kEOH*UQ)HT8t+
zF<5V;E`XTMxO|omT@P=mtLst=!vr!<o;;~NaqDB32QKU5&#tZjAYSvYi*<3|{Z}Sy
z{09W4BGkSY>rOPr3NY!G*uAf*)T)g@`Kljuq#i$2S|eV;M9gzL-esJWTiS-pi!YaA
z2~q6Mxe`%&H@){7ecga|VRCYEHa5@KL6Spf`v@3tFgF~wVg&mQzl!V8HhkV?y!xwD
z!mer;|F%U*DdTOjM0dU=c_}+DtaB~h$e}Q@M9o}L(5;G-At-%(wZH$a(I!mtC}n({
z2P&sF_CZUu=s|+cM2;&)KK90JesJZmJ;F4^BI5e3lGvJtR7%iSjvu@X^-cm`2?ei@
zzMEPdbc-^0#zA<XH+%K`E7H>Rqf4t**9;!AsHgkhCVMwwuT=@AQ@&Z_3nK0m?=Pfs
zGALt`MG3@=PC4bDwd&rTnrq9KS0_RT3HBbhy6tTe>@uiw0#VLNYavP`Q%q0M0P#|M
z>xtMK_0!8)9U;^!Qb}Ly>Ml1>o4nREVCBl9th-NQW=J~fV)<x|w1_H0p<y=^e~f0;
zaDJ`By?gJ*@B^<6&PGd>`S;)s-pXL95?>)7IHd{OqQWpAr|l1~qg)B)ES3DkdnhE7
zR+T>%D$ez=(+`F<9Sl(%eN=IfAqbO{EiNLqgv5ed^?(33CfF6jK%f+CucCRPRWjvY
zdB}`>;xl)1)1e9#^yc6Xbxb~TtT3$15go8ICfvh0N_NK>#kXb%way&oC^R-plv(Y*
zgYyc!dIGf(!;}a8b%sJ8Zzfk83x65=#AX8u_2<unzh;Y5PdqRiE*TVzR?!-6A9F0n
zX{RkUo6+^nxH^40NT<~q$ClK%qgVKhm9Y4?&4}W?AA0f6d-2-aLeO0TTS-;4grHN5
z*QZm6eqS+cKkOAs$T3pCM{p1txLo3Twl|XsT~C=&t?IXPI+I6Vv2$=J%5e2k=Zoo<
zAv&N2K72jet?KuqOGa01%LYx$1~L0jV_rG*g2wmLaZ6Z5@{VY+f57cC?@R4O1n+L{
zezJkSJGyaO&F;|!$da&OV1pWYpT!<+0S~cqt>;X#66KvfHgz~qX)SuS$HY52-{*Hz
zUEO1^w@fLA`HTkaGK|~{g(mq7$FGLES(&8!>Rg;`KHPbvvGi=zbq?AlyqUb74cZ{$
zaPy^fO5>x|HB(GCMV8h^;_11v^772m058h^bV{Zo&yJvm6G`S5Hfma|KCBkF-a3oU
z%)clpnOT~+q?o-)X`(c3RdnKcEP@(LXWSL0aFwaf3Mtszg&|8Z!Cr?#b}A_;b(0tl
z9H<|2oN-GY2zV@c;*j)>MOuABL&IJQyDyq`-)YtgFNlxs5s`c6>TOm6I=3=qxzw8&
z%X_AqA~n5t@p3RUSQMQ|QqygmcdLp9Def%AmKWBi#1N!FeOq4nKBB6cU7jlnZ+w)`
zAVl$M#5G=;aqEoeoErnHzrOaUl-PZ!(Hxjr^^nQGIojlv7c_@esj`FJJ5LlvW@1Ai
zUTM_9A3vk^yBN)6CwgY!<r;6F6)`w^EuC7UEq4C=ZPK$Lg4=y_b4Aq)JE-U5wD_MV
zCv!b(Foo~c8Xm598&^hxXJ;R}vVRvW>XgES-PQ3l=$z+k!ro~dn2})Lvy<P?{VJJQ
zpig<I$a~~!jfeT#6^zwjd8e;=&1yW^nBa5*;=eH1-=^2s-2DBx<n5cVWakDZZtz6^
zr0y=!?Rrvj|J53)IQf#I4##x(9CJ=q31iNO5#rsqmu-Z<hq}>U7%|DU<i02^Z8Ou?
zJww8(6y>4&9OAZh9YU|yIM*o2ntYbM9}p1mYw(qR)I(A<EI@AGLk0#09K(gxz?<6l
ze^VoB(LJ$@#{j5tI|P%|B*Jri`ekKhwbgHAXRS71l5Mr57r#$0YvS(^WMS7^|H_qZ
zIYL)vts7i|UmFcz%vRVD?&ccbd^%A@d(Z+_$6Y(C6F<54c{DTsIj2t!D-tj52Ga;3
zlbFP#I(S@y+f3xUh>mYh`c)_*8{WdYBht{2;G%n{wSXtg)FVXEOrc@A7>h11FVC7-
z#Wx9hU%azMJVi;|#UJDmLhNMtnCp%yJ7ULf^(RZ78z8lAZGG1^*j8$ExHD(+J?K-0
z=GocCEJNg0P7Z>Sm<?PDnY4;!jEQ-Gacci2SEw`edGW%d@y5ODRKJTy>|T~0F1E{M
zI*Hz174@bjYJ2isE|Ojein`1BG)u}a&mD<uo>TcO-2*MQQ<6Ovrx{p76w~HE9T(G+
z)$skua<?zIZHql5h6tGmd>8gF)~_T-L&k2IeRe2fs$s@V3AFHXnhkbKc?zXSYufqk
zey8rkcCP3yhSC<g^UIjyaSjIEA0cc&!`9c!D{I8lzum04<P5bXVnOFdZ@R2duzx)1
z@?BYnw2@72JDLp6<X(C<>H!JolyZ#{{t)p2lh-Bk{kZ$333zmJs_$yB=`*~qkxkLd
zJ7Rr)?mut`^}h!Ou03q_)0rSqFsaU%e5GTAz>#3c$N8+*#~!Ymu`2V&T6@G6`&2+#
z>et|v@Th{-H=Q{jJ}bgvCot&-+te6Gu>PW(X>N4LupShId%Cq2VJwWjoA3UyJk~#b
zanT7n9N$Qjl<W-OE<A4-Q<bmpGIn7q-Po7Zi_f!)8ZfR{%6VqiK0P}6PKrZiQ#V=h
z;k4I7jKP#eV|4->)uN=uI=xZBwqR+@Ht`t1-wXPaX!(<PldwK6v&=W*k*oH6i)1f~
zVE=>e!6Q6pjfo*{VjR<K1m?EW1NnX27Y0mhUHmnp=0v>3>{d4lEUiprF<vh%EG;qE
zN9nz-@z(~s0v)~ye0>m~!s1<HEXl#Jx>Hg5#|}N|hc`1cKNv;ZtWYN>T5S#|W;Vf+
za7Cqm{HTnhOE;>u9izEt2U7i&{WRYp9g*6*(zADNaShL44vK^;S4AXy;j%(|OLT3T
z7WPM5*$~ARVNc+;xD2k_iZLk*mFe_r@VZ18ADtjJi66DItSPje!m7|)GB-Z?Fc4C<
zZog=77@w69apl&)2tDR&YdquGfte-6!;+osnj^;7D9pN}MndsIfqH^Hj%lT5UTfeR
zy<}Sjw|#0}OT_$eIP#NEV1V?YYc{m8V3X#V+q_2le!htDhsBI93zM-Xd8_eMIgw{-
zRZ&MNECg|;jFkw%vL3mcfs#|P+&7ESX9?n2(BsG@5Q%UG166O4V|q3P17mi;5tT@M
zu$iSyUs0(J&-(G`e#QFz1d%&G65>3KK4nv*FNw=2-xv~W2L8n0IBAqM5c=S0f)OF2
z)@1W|@qJJHIRR)e<If7=S5*oue#Myn(zMK^V3gNNUf8&cOGP{f@8{#gk$PLDETXuf
zwh?PU88(|yAxD_d?cNQzU_F9z5PMjb(0A4*h`XhFewpkt*%6^8)f-2T{eh0otkv~X
zsMF^gNeizm<t<d%*CZ=ktRURz6=qsFi8oUudw3?aEJnnAw2OM|CmmO&yXb!8dzld#
zKMBO3GVgoPi(of<mxM{Srs{r)b9to1T|JQh6>k<Bi9D;=bhSHmu`{``_Qr?uTj?Hb
zgaGqzTU&E(-VLiq`O+pEZk4ni#OG2~tx1;#45YfP1<X>?^UQS}QcfvUm+6o<LPGW$
z`J|nH-Z_T1;GPbjXD-t$q2{@v`>p&<AxMoy@(syb>77KobP9V>ufM>nZMD`x&Gg<K
zR0kz*Z<YbB0hVhTxeR7id!AoKM3499I&P-UH;O!|RS~F`{!n(!Lw2XRrgLFCi-?HS
zr<$u?eKkMrU`iQS_$jA<GVrqF^fj%lffvzBo+rCmiPx?wu7cElEx!GlEv(5bFmOl{
zFOs)Ob~`;!pI;Lo!DPf2cs*pFe%*$^S%UekwBXES;2yk(_8Pl9q%o7bsBC2Gz%iiw
z&uNdKwf}F2JpO+?FY-TJ%FTfDa*^KMyY)Hf^}}w&sne+gH!=>v_uyVk#+iLOIy#>$
zJGgQbQcsgM@CEmYBQ-ifS5)miJ&$p6a$4Eg6iKmIn+XUA^woGn8-I*R)kArPxoN-s
z_job@G8@rRTb$}IvV^W_pdQnSS!ZhTr`9ATZE0~2?rpgzed1V#=3YBHI|IMS@$H}9
z3&eS!!8kwVoO?Z&%~8x}UN*&Wy{vb;D%z)}vU@OuT;3Eye<t?EnImDy7}PXaa`W@`
zId_*lj*UH>qwwy?r^?Fg$l&AI4_sXjXK1$I&bzU|iJaYMLcatYyZHXhK1K`P=Y(hG
zpORkCC#SrQ`uAwg+mqMKQO(>j;cIJdNAgdcg_-FDT|QSAe#5qV!;LV|<Yp#%Cu$`<
zJstHpD=X{Vc0QQ@0D+Q8YU@6==eFerj+F=S+`UzKefK}pYU@@5G6&ykZzWT`<yiKC
zl-PlSM&hr{&8{gSP$Hgilr(hh2g1DQQH`@Hkt@>nXnFPRv1ZoJYZ{+ydWDcX{^88N
z&?X-5jLcM+;Mxrhg`Aw6IlFg(I=Z@@BO~^nS5F<{32OLr{d)VdoR&x|awH~Tkf0=$
zq{Mp&MAPG8d*mnZ^Wz6I^|k;eg?7-Q3VQcPH5Wa=Yu#LapO0E1%P1);yKc9&we3Jb
zm-cxnsq8J2&5yozE-oxB=AX?f>hFD;$18r;vwdp*sc&bC8LXQvA8-1xc6E>r&}nKY
z6hBWk6bjPf-&Ho4Cnq5)nzkm`G~rZkC<H_tRR^l#Q3c1N3d05>{qw7!P*~a61{yC$
zcWH%mA4RV>g-%10>wIBpD<cZZCN3Sj0<J*2|EiN{LiZ{4`g^xk+y59pR?e^)0rPYA
zMjvl)E9qclV90LK^c23s6Akk$&{L~$kur*Ymynh9H3=%DH%K1P5r{NVv93a?3+yo*
zN9L{50e=oAZDkYb@;h4IynU+(hlpgjl>Pm;dkPNPN$a@B-?6z@xc6m~sV{Fp(`gkY
zG5YP?*q>k{h~ElDQDx31u+epObyX${7G0dYo5~s*-(t>hEbX;&232QKv_tP^uA-(3
zoO=l+*=)Z{OLQ_dHSN0pAt^O4uWc!brDXT%VcvkhK?vct7ALQKjE(K8B!?1wpJ!F%
ziOwV&ix__LVNFeVgNwf;BJu=IQ~8WDr_@15@W#eQE<=dQN7zMrf)M!^FF~+WR$9BV
zcGJL4&1HC^Ux&k4-}&L6I#pU(IZ(Vj>*eM3eoFdbg`v=o_VykhBl>N2R@Oc%4}8ys
zknW)&IBY`?4VJdz>~Mw(!*w>@tnjr|xB4wBX^(}`!1$Ze%wei8=#L~GzY8mljQhee
zIfV842h|S<L-4$vdV5cy-jAQ6mX?+R{IqY{US8Qu8%XXs86`t=FHcgjq*r2meQ~0)
zS;uY1Ate0~c>fpyv>+2}%3i(3?#w+f8>Ywu0r3<<6~b1At3X=a+R3T&<qcnbRWbGr
zq-Q~fD@v;VR`=5=MIQF-&(khaJ1*x9O)r!6te*BnOmtu04TBVfK^_wnzasi3wNFu&
zPwnkvEHS7Tzw1*^;R)+w!H=c~tcRg7g|qbNzUSvb$K?2e;R1%q{c$!e>pO2MxPL14
z+aB4*n{0Bb`~8VG>jnQA*>!jK+BV$zH#EWp%*X1?nE%AYk$-N=u{J|nvnt4{=BWSe
zHZ5#6`uuz1Z~WPD9@>9*55!hITqnIhOx0Fd<`-O*lcmM%m+IdvhonUj?lQ^f9mrge
zHx<H`Ljn@sG<E;~@uN7-^3Y0dz5csV|M@5g;~%{rK5-vJy=b#2bev=xfnIF`&V9&&
zXxyoMh~=8YHiSJ#dB1lVacMB#98jg?F$BEnixtrQR9#J<*k<8+rn@mNK0dm2(sAbJ
z2_!d3$je6!+E>A?$!vL0bQcu;C%PD%zMkG?UEQNlk<DXx&^#)6iiZ~7=0q5{nA7m?
zw65Th!s23cBwB?`YHeNjbkoGe#p#K}vGKF)%Vo6@{rCK&lNERL4kII@K6FZhd{!-B
zOIlf3FMQT6M?LNqD6uvr%X;U+7>DWu=q_z`nG1?92HJ}6zoy}_r0@rC4-_ijn6TdP
zUXQOHhWCYpEH=aSbb_v`+5$R2|0i46%i9W&e1!GR#j$}*L&^sEMpd~C&)@VF_^k@e
zVbT6w-onz1BZ0nj{ock0n9oK$-`?=&>F`G1=4>)!3Qb$!x}v3<rS%1e<A1~vs&Jrs
zYXk3LM>Iy}T2@^h-W16&axxgF59kV47N;i1m-#F(mrr6h9TF>l5T3$0us>h);}2YU
zXvHb~jCSe-0%5L%_v~9%HBmR>NZZI6Yb)ptu;n{F5uN-EHUCcKNR0jefQi*-c@Tti
z2vyPiuSCGFXP&@z*U0EvmeyyfaL(r;EYUk))c2GQoJAn<y6N9U9x%-}WfPj}3)OaL
zUaFjia}c?E1OJTdJ(!)qXWmLWT_ySa;|fDoM>@vk(q*?9gBlK?`Qlsai6q@>3AP9x
zt|(d*2ZQo84uhs8uB?0K^`aGb)XmJCfo@Rp4kiqsRy)PxwgQCqzi9F)yHC;cDlgcx
zfE9l|0<eI<nMPgy4~kuu<oNpS+n!B7|6wB2b*&0$%m<g#kMUESUPAq^lfKZPFPgCP
z9r4MV*RL<1j;S}<Y%-|><BoU$&e(X%a4`l31??~-c0z*|ZV9P9Aq^F3ADQ9YfY!F-
zKnK&YCfZ{y<L^!Wb7KP!6B_ZWG<tZu?~{*k%i!8&Y3+M*`LeIQx%o~={?Muh#u5M-
znCqy=!1=(9MI8K&hat)5z;IiW*}Le$0nRM$exab4-bSkroOMH8E{onV+Lf|(0n&Ik
zueF7RQoQ1h%iy}fiHG@1(TadlsZ~`5zG}Q0<IZ!lX{%=c?B6heAtc4o3xulvg+N_i
z`Ef%7%I4_k@juAQ0lY#WklasLgVTfSwvrryt&`av=ML_<XJw^QCK%LEn7aD%!NVvm
z@x)yyXrmckh7Nj3=-$}Sb2+-3W(PE@pM(}8NY!9l!|ZSrWQ-j^39>p;<b2tXzR6kG
zB!{ur|0KXa3AdE_r3QbjWbRwG&*Phl-dy=5<>h)}XMg=-*w4awMyXc$rnLGEi&~6@
z)kdvZbf;938hi|`-p%U9-Iz5YPFb(w@nyA-cU0fN$=mkOj)O=#iY7xzpu0j9PWgbL
zG29B`UKwTs$JBc9m;)Dw_~TA%1~G8b+)n5NiU5so`b>v3jQ;mjX{ucqVhzYg&nF7U
zC&930-<p5&WCzqURPXAXgiAn61tUsJYjl15D0<JmytI@V)f=Z{aE>+LF}8b;ACwin
zV_;y=HTOQV!s(IZAEt^ho|>APhwU1T<O=bGH%Up_(^sBfJz;bpAOdP6Dj&NNKMSS{
zQXjxPff%A}Cfl3*6K?HBAq`t7o7lC-CC_YtQ9)O$Zi*$(HH-ggb-T^V2FDyFT^@Y<
z`0*pgjA?-1w@&|(i$+VwnfTV{ACtXq_;Q3b0dTqOI#U`td0*W}&ZL2T37S$sb$_%V
zJM$DF4DAnp0p5(j$Av+13;4>T(D8BV*R_ZMDE)*|u9AJ^pjv{`6L6Q4lsuztZuVDF
zQn~7?&6w?O6xq3knHSyN?ArDWUuMh9H!Iw8O`}gOGhZ|783v<0*I(3(yPerm0|~FM
z%Dn(%dOOCz*cfP}au@wz+Enr9&rgl2-0U`P|1^+W-#fJAuvMmNH04%bbOq2!E$P3Z
zQ+h(E>Kx1e0T6cHYC7$<4`KHI>s%3y(4qdL`QH=t?a@4<k0+x3J2}v^{o?a~@KOJF
zm}|EW9szJXu`NwY@DDiM&J3Ac_B5<=iy%D10iYo#RW~Srq|e_#eS6vwj;|oVAOWXS
z4IF9?Q`QK7wA1qP@~3{=860k)d$P2oWXIcq+lQI6B;f9YZecdhwt4ViSCd&Zomh5`
z7XR2li4F7$qA{trW2$_}OCM5Jey%ygsW9lJiE_S|nGZGo$Vi4G(xOaojr0FJh2U;=
z@fGtrK=T(J(+IE+EiH+D?$3o)+y6QxrA%t4^^ShqxaZe1Q=dQI1+6kU4SgE$NP=Sk
z{^2&g`FDY6R+tqAXJnj$2^?`GFFp99wu*}zLkqis0^V!Vi}|l79yq@r0h-~oitZSz
zdqRsKZmynS8c^2s^aco;kh7srLC*Xqgdj{Jnrt#tCnq1cxVS8QSkUVI{(T>au9DO=
zUOS4E<0w{ATqK(i@j&)MzM^scMIEUV$Ws?MGaLM9na|Qn7DpkfsnH47{UYrQ$450y
zy~~fyK^QJs<aGZ8WOf*_MR>*k3l(@!(HuqOR~H=^9Q3sGeEH?eMc(_Q`C-8eR}R2%
z!U&D#|8wME<9IdUaQ*^?s%_%@=V=FcV>jUP083D|0%1`ZuTabdN;fp^9$-0gM{a4b
zu!LPhBppcSx!5#B()ABsK9HB6pH@>d{OZ|vB59(Yj=K?R6dQ(sgXdYbxanvhgSnwD
z3&nR$?q&$01&(XAE<5fFtd?YCT+qAlx%BC5mJM{h0-*Gp+$=u(FX#LV^bNVo%y5tq
ziay?eF;I>ib7bX;QjV7-Q~f+-{xRr5=<u*bh;Qaw6BoT|0ON3<t!p~y4ZJFft4Y^a
zv}%^V3*elYnVG+ti{kD_H3z`yD->XSX60KrCs{UVwCdeRLDhl(4@cUlLW`_=H^-mW
zznj1G=UEro!biePy}Z~UEiZZ)lJsS`f?6jvz26~)U?|-v@;re`0nC#Az=13x_pxL3
zD=RL;#GAZ_wD12dWb39BWlQY)lxdAkRR}_%pvkk6cOMEw=+xAdOUJ30>!><uIK&nD
zK=szG87`>*LXUE0!xqSzje$RBKVU}ehO`4&u*#Qmqu7uv{2!?oc&tIbVeFGo2-BmA
zw{H~p!aWA;U}a^!4C>j#AG+#?iJ^)-hq1A(lcSCoIt~Pst*E5jI_+9>?dePgoXYB2
zrB_u`gJz5qDIL1i*jX1El(E<Bso^EDgz)^Jhag9xehO#C5dHx5^u$Csa)usY{#v13
zY%KF4T2O7{6r!vVon`APopn)^9(CmCQCC^D2p|{@6N}wGvTPALzOwPHP-Z^}+6!n&
zoj%?5lHcw0ahsy8-qHW)cf9PcLqL>}qscZkYG7Qja&l^WmjewOSiT@F9R~iz#aSNx
z0!Zg(g;qewf3sYx$KkS(lmKQ>A(9fnY<Qb;6A8)EMjt7W<}96YD4&A_u&$MZrHYPb
za*fpDkNs>L*$s#abn4EXJC)nO8(y@SpNAyZ(sCW}Iql;|I`~E6o8N%fCDXIdj~ds_
z%?UonHPD9qD{Y$odOU%#(|k}sIlSu((5_1tL$Z4VR=V$~V|<~=qo`Z&0DlsaL=!Ut
z)~_+XnsNGdB=yCKuR7NIwO%ndP>emsb`85D9D=FyD-fYjMgG_uroh+gRz;_Pin&Uw
zDLQD8?OwqdGpMqG1S}`lJ3NMyG%#eB<yYA?%7H4O5J)`H!S-Z03rX`y++GxxGN$Ex
zNZNr3N!LHJ?Z3Df@!}LzY3X;5J*qf={`_Mw8QGKhVu0_$p4I=wky#xD_1Oy--XAFg
zR&D*oLi}dr3Qx=4oix^8XW=;rzzV=Zs@}lm{JP!&4^a_f>N>6Qxb{<{ENQ6d*jN|C
zSKy@oi!A)~ISCLO=^`C^akN~H#?H&DGVQk~RON|RGF0DL7Y-6CH^F`w&&_`F^TXWZ
zz1-u*+8@<Z0^2NdHj-_{!=7V(<8-650K&my(xXYjX|f=|LSeSHdhp<RjV6*UvomTA
zbhG{Smugd=WkZhL-LXkg4Y&3lT#E-iFZZ$Qu2=}q%clP*;eQM-8F>+GX{{l@SK4Dd
zDgW<dZ3Z&!l2jf}-(a%a9eIZj(dM5ebEE!X|LB7H@}i2mIm-PuZ`gH-{}8MILZi>n
zxZ>_tH%>u%M9}ghSdhB909yKW-@k`qPC$4X8BJ~;wYG9~ja{C-^;I$S8|okB|M`zi
zxVL4L976uDC+b{+-0!$4@|P3+Giso(ucMT`B8=#yItwkMFs6UFXM%n26SW)#tFw0d
z=(K21pZf(_u3dp{ZU9^6Dmd{AO4ifaGTR2cPf!HFLO+A|`3Ug4oKQN5h*3lF#B1v6
zjNa$7@&QO)TwPO(Vp?9@l~fD)_89ODsKFn(d^qfy;JI^vswIzG<An`4X-XagwCaRs
z1j4Bnfv<1E|C=$1=;<{g)&be1YxBdcreAoUH*BaVIXDQfIWkCGbX*v{!mxee*tAN`
z6X(Cr7*aU`Uh^@>$?u0@&;3V?+KSsC!h32OD8Li_%i@Vm$PGn^8)){_+n*4Ds6e?8
zw~ZF&@$XVa9M(B#6t!W#)UTEq35ac2>$2C|+Y1rML%%xwBR31EZ;<y_RhI`FNnoqb
z&8zo%Ha=<hpM-1KW!*mW$t<NsfBR@fNF16f>Q{2WO|#e){7jZ>rqZdEh{>FjvwnE`
z!fjf+0(7JaVHg?T?;EYH<wKo?=j=-dO1+OVpB)hQ5LYq3>@BU_pm`s}@!<G(5j$Bm
zdcI>-?!=KJ*IDe&9Z5XT9krskd*QOu=hrSV^q@-+(^xXM!)Nosp<}oCV?FLDTB~TV
zjIcz%|MVEu0wJ_3Oz_T0L2*;L|3(dNWV}Q&AYx1-bpQ651)k)Ve?zIe*!evqkC!dx
zm|i%W58O^QGB_CQCkDiRLWU%r(Fb;7!^%qWhUCFLb@pW^qgCK(0b11Xj_C&IA8A_t
zWB78QMja&`#~VXJr(2=qcOd^s3BpUydA=u6sFJ=L7i`57F{wWpikKP9i!c2%r9$R`
z>(GS)*B0_zqEPn)HGpaM<>${A^Hcx=x2hWJLx({c+{ox|JOh3+TLe&gL#lc!e)Y@5
zqHTX=!gIt0T0#E+|1&G=?5cd(-~ki@LFZW;|0?v2EvUZ?JA(~?pv87qZTz=0|Jhmy
zlbCyE#teW1s0bb&s=M(lv=~%N0G>NJ74n=oG`=W_bT2eRJzOP2Qya^^?>UsC_z#un
zG=g(wv!R%s42Q0vcMcA&(jcz$zt$t%(QCR3&>Y;x5L3$ZpO1vy{8h~gN@=w4p|Igl
z&;*BiXC2jr+yV{$qkDF`glK^E+o0jeJnAA90hWOAdk0ro)Y=c+1Jfi*jvNu5Lew$M
z6r$(|XAF@e;?Vb9x4IJ#&yGR8W1#)}Kn^(2P`|M7>5{-@#5i(2Ku>@C;GWEBU$~{X
zw;6hZyQ_wUFahtA7`5&0BHF*g&{~)^T0=EOCDyrWn3bKK7E#n7`AW_--4w3biSi(N
zWA~%}uhQNG9Lu$h8h)B*Wokqvky1*TGBr>rV=`rkC>bMUPPGYz3{eRonTbfIG)qD<
z6G9S_BqY<nt{c_f@4Mgc|NpPUu@A>ip8L6<>%Pu&t#hq)p*^B)bc{g#E}kEJ|9a?F
z`A`8Q|41+8au2v1s%W~0>Bf?I3OhsHG3MRCN$*>LI&wRR`{Q0uL(d^K8ZaI{OQ=Hy
z&KmMOp>E*4J=^xbqIwGZHs*$|$U*Mw2idP5jD$v8kKW78%`ItO(EYso_wSpX-5OQ=
z_;v`HbX*R7Bj3{dAo)Gk9|jP?$xn39PHqX_jySWUM``#ZiJt2;=JLl=y;Lh}YYQ*g
zahhD$y;Iu07uhHHjF9tF*)G)+u>8fD@1K<VGe}y-8|DNE`Ri2GR=?Ex6MWaqd(LS9
z`UXhbeQ$5BjzuLqvnF2PaJyUvAD&a)A`WHR*A9iVD<5eNrd(UR-hI$h&j~V%&S5F#
z=%hNqojW!A##AJ||3r4SdqgauYMQRsVV$kNTz5-wfUFBfDVQH~_gnHTP54xZNmE7W
zcJn#pvk#KJjVH6xyeY-G+oZEk-%eLoH~h=KJhWZRgR>At*12^zsZVKP_1%P3!UzLj
z=N}(^J`$_7MJ+>6_Vqai&?#l*<xyJ>hz!c+Btij~bYyKEeO;j0DSK^|*7Bv1TQ`Oj
zw|v{{YiY_|x!z^VO)e?#K-p;4Z7Jsj+)4KqV{7&PoF0qa`zB4+7xy7$ep6QV__PM2
z*`>{!<NFI$E!S%Daw>9|<lOxJrfO=Sa_?Sw=MvkSL1TuHGTW2{6ttEt>RMMf|Ia%-
z#RQvluyG@N2Tw9Kc-Frr86utuPKFvF5yEu?Q_UvBw4g7OKrn63fkA2d>x44Lyi|m1
z*gp_xkpgp$gXSe=(e7|1zI65~w7bp4=Q@Ya|M|osqPA2$=Z~$t0l%y56u7_@^#T3y
zx&CntWdB7m<88Tq88xNOLhP@+&r73sqJk>%EJZy5q2Qq(;hHDdXas8cLo(Ftc?}KJ
zWD!-%+S=Cq(ktg2^W%;&-BEJQZ8kL2E4RoEAbv9@=ZzQw)?`|QliIxLs_X;X?v+%L
z#4iK2d^tIJ3#of&rjT*;|1C0JONtEg`u^Y1*#H&fmQa8AJnofwECjySpEk|d2=mjM
z7Bb4_Io$%}w!eXVTg-7lognNBa~4Muh)hi_SC0R9SKYf^Re}7D_^#wRxi$B~GwqBx
zLv#5rbzGA^Inr{`@`E~`e<uc_icE$_-W6TA?-RKJw}odnn)bxRuk-o%)3dX^0I<pI
zsZ&v<=cgG4Y|6z3dU}LgKNx%P8EK9-cujBI=DJdeX%;mzjh#j2qojsT%o`@B*9D<V
zRy@jiMa@X*y4G>CRP84Ob%V;&pRvNJHGd8Lw<VLML@gPD&qLUi_j*O1y17$E<L~%l
zGY`)##SQALFMQrIiTwjdhkO&80Gc!!1vH!GG`4iJ{CXU)yD7(2C|By|>7@L$06_*d
zO}`~YQ@_oVpbD@W4;NK}V~M!>J>+nJFJXdSgP$ec_MFR8<}~p_Su)`BBw)qc+ssH9
zKps1H$c<|hkg$QwB3Kmg7<@E;mQ1uV%I>1z04T1ZlI1wSp@97p^^Um+a1M;Hh=3=%
z$&SP9hr>R2gvG1r*h@K^gXTSB(*pA*Sf0*S4D<S%%rjRZQ6!r!1utAyR_0YoU-*6s
zUNAnD*|&)~1Qj&<G$g2QP(%{Mf8lBK6&)V}mE)1@mz@aCxxul#W5=DC7%0C5PHxu3
zpzd1AdY|uv9laYg^JKsVnvlzg5XgL5;-ZFfFQj;MbUcQNh}-h*K=sBsd}(ZxlLXDU
zP^fw!xh0Gs4P_jgEA6PQ?NVa3k2ni@!8t`WBP<%bWB*Na2M5hy>uwd6suK_I$ljK;
zP!hH@Y$XLXd2kQlH&O_JxNiaVVe;DTHob}=L7L+{wyBxZrvdNKbw@4YCMz~N`bB)e
z5E|@Mb%PWpelUhlHoc?>8*o3B49og3IX`E#BcAgB7>e%Qpj)N31RuKC7F~XF=7q;M
z3IkP9#x`DkTb%h;JipvA$~8luabi(dFzg>fBSR*@o0%+Ok8bF~8|lnHmbx4)z?Xg}
z@%;YXZ+(8v@_=BBqV*Y}um(yrPGg$yhDHP3**r4kV2A*3jERT#VpV?j=5KbQ%z5&l
z_USGjD;V5|5uKhJZbE$jQcqb&tQ>xc%)578Q7&$zY^NUoI!!3OZF95*c`QJ%Lj9ek
zWv*T56Z<ZvyY{VDK`Z>?tLB9mHQd<fXM7v>3f!|$#%RROOsm*pqQgtS&c9N&wW4t6
z6n08iPW5zvZXUvE8&B0Fdxb~+9Pw3e@~w9+qTO6DKAV)-L#!Pp#eSVgr$nBO51SaI
zb;`YL;iYq8%IDF@fag4uv$HeE|B9d&av}w`LRb8zdd#wo-3CyvL*T6TMRcWi{NA2t
zT`d`s>`FF`jHmKO?DK{W;Z}l(VW}FNP1m(u_t1Lf3=c`^?>kO1t~X=g=jV?;kakV7
z>^MRMf1zcvoblLzBZq-t_FdKIJ<mKDX;lnK;Gq)gV)9m`+%J%I5?{WsGblw{XG?H`
zNXD@kJ?x+v+&6_6(_mu#QPSEf+8dGmCwKrVO#OFZiriIcZ>QezJtqVm&)+B<84<;R
zJ3Bi&5kpBM?}z%-#vIA{q^MJJcKRzJn=~@e<-xYLqCUE2=D#Bf!B<m=LXl@FVN!2K
zZUC6b*tyVn8i+h18aWDFDtIBr9cgfgxJsluwqFl3Ti3`9qSutL0p3#CNPr)05~EDk
z$iD$TT0vX@HE}01z=<OJ-*;DPe~#8WGqRJo)JeU&58fS>)p>Y8gLOWCD!v7MYk79h
zBJGDx*9A*&yJOt?rna_LWnN`+LvFL(J&lQZE#pSUUpHHkX%@yDgrwxbL<oGdDVK^G
z3`=ecmc3~L3`K>VOwgpN7MP@Wh&4jEtW9xgDTK$pEtfrSw7-QR4S$p_y?-OlJ_EP@
z?%8n{3W}cT(HYS652R+tOJZjslSaRkjV?OsCxJjLby!#3>!{C11JbGgBWt~sy6q4o
zp{BX*c|0Zg;o;>amrfdrWOW|>lP4htik+qP(5bc)uL<-$sh`LKUygYOA-sk5%dQ*9
z7+>9<uc)1pKeiK#4HBpfR8S--wd4qn1`yJHIQZ<)vXS>RN8Y%H)jE!g1mA7FNG2T}
zgYQQ3A2QB3(~uqAv^za?tKZuP*GlvTzZ@XfaPT0vsmU>sh@U{c-@3u~l5e(X8}JHN
zo~Ro8pxZE7wRxPb4v+Z+AVt=e2513~<Jp#+{JfKE#crKWUvIFC!LB%!79ksgXThp)
z^Q_s$hiBn2dr`$)!fO0E<2|n&M-AI2d6CIRjk>vf9^eWVlB$E5tWQHumSgk_p=~z}
zJ~!rtoVJc5?M_n=OGiK<2M{j=B2MC~X=^4Xgj2QHYtX9z95M!pM;NL>t$L-VrpoCG
zbT1@K=A_DH$6k>M<B;kkd)=Q3Ez*|Y86@-v9I+Y?#-?O?to`uE!68Sy%Iw**1x)zw
z^HUeVq0Bz-)$x?3RRC`Cqk}LmS4wsw^=lT5MCT$ws>$mVO-rFBe)O)V!ByGy8@$rm
z2EW~^`kQb{XZIaKucty(RkY<J>Ib;gghC=3!lekY%%)-VjQWS&eg+Ll;Ipj=$Y|>#
z@M)bVka*z^9eh~$1Vav18oGK!v#D~JsWd!!G93&1n&Ne%16wQs0Z9y2Kq2nZ42t#b
z?>FF!8*{|Mq!GQzkrzPOn2<qu7~$xGu7K#PRJ0$eCl6$xE1;RkXl-3S^X|Zf&EqV7
zQ(_iEX}MQ+bHIqwd@8o^vh<a=9Xl}bg{bz+o{XZlC!ar?K5g@wDR;9AlX~sx0$aDz
z09P}Rn>KHgsfKQE=$XcO?;R9WFpB^T`nk7UeDN;9K!y6~5RkOO!f_7;p*4NDf8ny%
z?R~~+w-0wyG`&C6X@p6La76if)Gx`!Lf7SRxuw_c`K)ioXgL5?t><a3S)!=B%=5<U
zG>^o)hOP1>I&BSni1>WmijU8@8>9u%QN}8@C4Esd5i$nDH?m#vStsPFBC}u`?dHT>
znKfAuwqy(u5M3BowB&RecsBsJ=Dwkd2xlnsx+)j8R>Tbv6->)raeqCVR8N#-bQAr1
z&Sb20c+_vN3}cbLx-QgTgabHmb7r?_9N1~Pt^h(j)Tgb>`=ke9ieipI6D~2fG#?WT
z^g`3_3P{KnwOLJbz2WMarHWDQqkf$YATyK()faZGt1~4<canV3vk{?p<COdsC?C_@
z@33k9ZjamOw8_Gi^ZyGAWYw0~Xt@d$W`r*89QQJAeYc^{td{rC>i~?Ejp^r=^?d*p
zfl!5Z!8*en^ay0m9Crg=_>S@z!TZxC5C4twL<I=^t!Xsw7<p$WO6OY;CC;{dxoc6|
zTkkDzS3)&L8lve{n#ZGbHl-%2Y%~=57;5xu9}L{g>SzVA4q{q30%RJ#=1nBtsfMmT
z?7QJ;1nUzn^l3C28ie_~bk{F=(3m5YPXCYAI=xa~$}rD(?`Ws<?AY8ZY?q#I#qfZU
zOWxG@vWOH0e<^{5_tR6<MGO>%ExH%SIJu~ZvCC{sc^#899a4WN2{rG%oS8rgUkgtI
z!|UrJWg7hx3}BmukWXD%9aUyjtmpWsE^uK^gHCVuCufG|JgymiJGZr6h^;j}=^H4Z
z_jc?^{fxun67g)7J>1>CwNQ2x){wp6_QoZ1-)FUl?ZxUA4~K*=zb;BDT}RdTP)rqR
z?=GR4cCUQSDf!$G_9*d~(ACUyXTI-usElxdK)d^JeAO{baw_b<X`)5jOvhQ$Z`|-M
zcq2hGb=X{|-&K9XKC*h;B5VZwaGB3{3yE!i>E>J9>RshI_tDbD-RtVir$8_nBVlRI
zGhQ^chyTf7;3?i0q4-`(^p7zq<9UyTlVJYWg~syV{TC$VD1F^G|KNYB$S#1?!^rzL
zD*u<ge#cm)MWg#9ELPok6{M)Tv}-5B-eIAR;FTIWIxnUKRbYj=XK^)JcNHJWcB<1P
zl!|~`YaopVbsYK^BEwRov_Y|jluAo2S!=gAhR6$zt%{0@(5<21Z9eMa(nu;Trr1k4
zBiu=8hZ=67I$kHkq?+Q&Q`MineM`};NPSA9n0-x6{pae&9usegdX1NBO0IG7-AYJU
zfVg++H5>=Ov^gE{x2iYGBqh?du~BE~`iWP71`G}Gnu+G&v)|hv7c54Y2fFJ|fLDI@
ztf9^R`Nd9P4<dGM%yT4jd94V=T_;Y&5Ek2Uw7(iy?A(BD>vlf6cc*5I^^JEu-t6IZ
zqj+MUf1t399uhEJt%rX#%#vRcQ%7~9&Ca){L=kk)Biwb)j@o?Av#mt8QMj3x7lXHB
z+Jk-@6L~ClOIBRA(|h5BCu|Z?^CmE^xErR&oZgq7nR&0;|MsbCTMLPdI@=-2=mcqi
zsR_wqgTGBkp8kAYpv&nuEn3-AQ0lfjlT@=Dz09l0*~k(%AB6h-=JWqYGXRb_*W*Om
zW>&(v#XEkc!v1+ihX6^BZjBWW!`4KF6uiz3q6F;K_IWz}O&Hs@wzhV*!mxu~Pkdm=
z3DZ$belfR&fk&Ke#fo$F-Hjuzo;P0p8j3TQ?q{dRR9PSMU}Iqz$QfuLSZ^T+bpmT6
zwvZeUuyVWK!e3(|bRwqE@WA1BAZ)}%0x6&f<-vcEhx?>?t80cNH}ZQafZr7~iN>o@
z7V)DK<y%0^JRW50>&*50e-a*$_FZgW<QZ;k*=<%4cx9Tk*uh7dDU3A0RqT#56N%4f
zM^nc6$J)n7!aLJYrY#XRfts#e8eHF9wcZ${c=jfamhF;VM7udPvN}>JPb%?~bgB~b
zP@tKpzD&Rg0PMU#b(XRMF9;sIVrCH}+C}>8hHRv@xiij}k+-i-?+gS`-7*X^6&Rpz
z@q~f1&%UE4<l412Vm=rUU%x)>ZqTVoDF48(a_w6zi(kO%g(`E%DIu2Z2uJLw;81C<
z*qK#0c7rP?Q`90$gq^NE*OJ#Jrj||(7(&9NJvlzTe|&*C1^OQ%Qc?_HogfiO@jZ6(
z81Qi(dOAeiEJ42Qm+kB)L2-BTVM1E*p@V?@Du)O#3Ka=kT5>!ulTDAh9fa{bEFx^#
zGSfA~URft4^Gog>ia6If?K26~^(_Jm$u<zQs6|&xYbK3F#P*=G{N+Mzb@i?3m(gaa
z^W1UV9A27`r`?vP*m9aYBqCR(!=8!8DsK1-!&Of8Jrko9hut}0DD_9T&h;&L%Enzc
zAdt6T_O(QO5ShK-s9L3BVRy?ka-T|aULI?XZFd^knef!5k@nV9Diz}UX>i|mog>n+
zjaNw#4ZpN6%9x>Y!#0Ll+)xWTM)8oLT1fz3?VZ&f!^6X6h>pf{asB!V?bLZ?COgp0
z`J_4{i>qSonU;Rzr_&SYi((JYCSmIU@SBhBPFEr?s4nSBc?}ZkJU5)(;*i)$-&Wdn
zRXQf=ImaufBUXjqi53t`7G7SdRG@QDoR>j_cnqGg%s<i%ilQzFtF$$h<Hd0hLNlVO
z0p6L5s!Q_O`5>p+t;7WVh&Hxlb)Gaeua!`l>JcJkg7P!#e4Y!A2w`j)?}YCL2??{%
zK~(`3tAj%}P~rrGVba8ea$5vnSrAJb-XxBbsQl`l+a_4UT)IMOOpM)g`Yo4WfkJ%-
zsa51dMx~Kn3a$iv;}NTAyfB$|_3J~)@ko<dVw;4lXiT0S@2Sy!IT%J^EfNi(;|0h*
z^6J{ksx~prKNxwi>dx0hN`ky?>FCCkHHWk*5$)YjMQvzfpf;$vhfypr7TJnZ0TU^>
z2nmvpkB?2ibglX4C%SAW`0g>jrf_CNLPAleqV2-4MGPw*4K19wzKi{Smy<RK?5;0`
zE;vhJ+lv1^7yOOlp`ZPCV)Ds<DX<g!=O2n?ABbxqKcW!3r9L!+6A#BnO1I1Q@^89b
z0@VJsnzM0x`#v1bgC6-lmae{9iVh~V^9dFOy4M};RI@7&X*3`>oTMD1{*n7MSfo3~
zfCy9d>MPBox!ZpDC{ffuu5aa+b@?STR#sjuZ6`YH`;74kND_JCwg6$%nPh-XvPt~b
z?Wfe#Fa((A^{YZnR*EVweSs7Zuxga#3`Md&=L~Ov+24xqXT4Geel%MjO%4ftT1o`w
z5gX=D3g}}2BiHb~M^^aD_CA7xiEXIXw2EMHe2CQK+f<<F2}i|}Dd>t80$e^dSKhSr
zlDeiQd3>h@^XlFwl9U4n4kRu=u!wp7EY-{WD?rCne=Z2^?_5N*`yW@0ua<usmCYn|
zE^<SzFnc9*C21(rU=$5)mf!jDv1jx|ZijnF5?$1K?K}q}!D|(FBU-&rpFXLpU3b!4
zZGO@_AtB|MV2zDDL(M_Bj<DdnFK6-2w`;J?gHqh!E1UAi*62m-Gl1X=;tox6+IrKb
zQ9I$Uc5Iu$rC2s8CA3+%ubMQXl9*f%knm470PT?2J@R}=WTl}2&NA~Cu=tTYT1pv*
z{()80?#eJz0BHht9XS$$4(Za-NteKc*<h>g0S;xf82q|jh(wM%6m#&hyKNdJ@31|X
zrLVafua9ZVCaKABfGajq03>m`i%%&w6hN+$Ab1QqoB5*3HlTolbXDqiL=ByO?$PnT
z^4oyLAfv5Ojhz%UV7s8_b;D)<+7Jpgt?oT|;BiXmGQNPi!0{5{Hs;hjFe68?hIiwq
zrWn=sYYl4gEG>8TecHTh15=&94F$3Az+m)knPrW*cNzqWU~GH6bT*jzUd4P4p{744
zT|CYcEsE)$3fw*w3{DCA+rJP20}mR^X}b`hfG7`2NCxHb75SKL@S&g1QV{EB0$ckf
z3p_e#-2T*}HUx~ldtJFUlJ!oT<89(M@&z*t-PAwCWCt(#-VcT$jIJvR{p)#y?(Y$r
z6XYlV#ZGLem~tYbfVPX>kLH>!qXL%|B-iR6L9izWZ)EdT<2X|6g)Q?1F(@RK8f<pH
zt-u+8(L%Nk%n0+aF_U>TTB2$Zeqmux6hJ3&i^DG~TF7gqde91MI=VT6iy~mA6<jPU
za)u0Rp+ILXIgwd!(-#xJnHQ&|ESbAzrFW+eYBV@m5UeQAx6}J#Zmb{+gj4CGNrxn?
zH1v<^Dq|+gD+oHu<b`|mn$Gt=c|a*)j=HJ7*^fqC{EanIQA+p_0kHWWGo*knvCGqe
z%Pl@n8QO)#;y;ce`LHkISsJ=A%y%1}C2C9}oj$Yr1A0hBg(MI90V0GUox-+i$*$jG
zp~AYaV%@#9-L^z6pGT;DXEVGpkiL;uX5JO}Txy=FCU=)2IGBi4uIU)kADYh>U!IsP
zUQ3DQ4IcRZoffOTYtO=2=p1RTn6f<2eGjw{yXE53b09Y%@n-kKdWe@!!vz)v3UftP
zy?yH$D1C(}yFTgU#k)wZN-b~SeuV|^UPHt=whE(h%<26Qva@5y%%f&@VX+nM&~{<Q
zQ!S}+8d|>8#BW3LZG(PRLM$HpqB>jgae=+wl5?HV5<;*_`835`OKS7|PBd_mG9i*D
zxniLMN0YM<q<_l8pYb>TqF?d5b<A=}T)Ip$<2mQED-ud;rn{QEU_|@`x4+`eD4ICr
z=yU{C&hLvN(k9`Xs|*3b7{zIcavz+Unr^;7Y!E*sLrV-*F>Q(N<!ybxXig%wwMny3
zU$64=KO96D=uLdk6un4oKeH|0&GVafvvk))!HQk_YlET+7P&K1enzr!kx3rg&3`Cu
zC!Hp{1VRjtMEryC{+l6c^3(f=;}W01f8n{Se>@#;JEM2678|FBhf#gj$(}{37vH2!
zG2wMg@;_*I1^f;vF@ZbQBs}*fwdAAr_SfeLR%W1GfDv);MLI%&XAKS6zq743Bc*2x
zgxWDN7xNaWRR7}bJX!<)Pnh(-@7ixFE1#!!R=fd>6h3(EHt{4){a`CgoZu;+i9_EC
z^?d_hysDfY6ijTY{HaiR5n_5$6>W-Qh{q*=z$Bmet3E(#fj*nZJL1?`uDZH99`AE-
zk5COZY+upa`o6Bq9b8a(m}e;pEV_Zi8?VCp_bNiD;!PLyQ^1XyE1&j^Kd-*q^y1Rq
zV$+i+=b;}s$}VMta>=R0IdQHN3g!hnjnJ2_1GYMjw_ZR5TWcbtT=R5d8T?`&`sV)X
zS4h!2r5--?U=Dz|%EsP19z<yUXTJpPxS88ZV*Q$wkxjr(*E0)m?ZbeNn2%7yA*ecb
z@}%do-q?se5sx4;`Bez7d}oP6Zw>Ebd<xE8GgB2^B>>-}#K=>Lzlx7TBwyz)6n>7<
z4H^%4HdC@!_#o&mCOkxP*3&kt#~_k(mwNAmh$g`$*~uOo6%p@P*+Ozcwl}?pa>8#F
z4jv@=lOd_GiweNRVRhZ#Ho408ZO;rqZG>8uzZP^ggxqsyPR|~EuD|VrRo0E76%zpy
zUxX>&Y!tN9S^#+=_oml_aVJZ7Yk0HdtJdD5NUC+_@ejsG1Z}+9cdbxyXU%R_Fo1Js
zU&wCrY~<+`#uytJe7og5CIn<Ss%lPt(_&2KFrT9##vAgn3($%|M|9ieBkVZnqoFWL
zkWL0Qep5cCtdbM5)Nu6vpX)Nv{!(gs{nVE_+(g6~&7HC)ru-x-llI#WwmF*|ys(pE
zo{Nd`_q4qcOEA|Z?wX$YvoD}t06LQQaF7x$Djd4cY|OS`!2%*w*kc$em&X)zkaLs!
zIIDyU3n|N+{5Q~;OQqQ9wOjf!Qk=~idU4u{>UTwWbwTSUIbKh{H)T|;Ozbv5)xk%g
zcb>_6ABlC4<yGS_YMX~>(S%POkzG^I@<QD$kbE;2Dnn@m&KK19wD%6YfuGLZn9yXI
zZ#g`X8tGSjd{Yo6mfZa|wGj@ONOwCkH1YcHC`%yClsDAEFd%|_Cem^ma?orF<tbBe
z0kpW^tzeCz80#B;rHu6a0gHS^=sg$H8s6+o!bjeKnS9xeqA3G!1O?(|4>`GLr#74}
z6_e%+z<TIBU97VFf~Lau(i68nl5ITU`bA^Jr7VO6`Sfj&zUDqE91L+9PLVe#A9OTb
zDVIGkAl2Ng3R}1Sn4aG5ZaoZTpF-;U({2-T(Fx=}9p{?C;nR29D39EdZ5^B|ir>j&
zWt-21ItgA_j7s)G;k03C*-Hn9yGD!1dMokfYr3Pu$1EmY>EKNI7Ofsj%^uP9-HN;X
zSmt5;AHS37l&R@LNDMj^pQ`CPU-e{4S}?Iv*Ju=Lnj$#`)BXD&yFeR2gfCce7IzII
zEg{Pm$;wjLgww7)+c{N8%sQ3ECZXrv!Sse7l^@j$ms}m^yMC}C;@qSDcxS8`+#}|c
z{8rYUts|q7WaFm_YjK@ChB<f990I+oRs&;W#bMns@BP`?uwPbPOG_nK>IG(T#3phO
zG#a)Aj#maDJMZ8@dJvhQyD(}M8(UTX^jXuVfve~cR)h;J_zGNdhd<x#0qyJ6gCiKr
z$vKrZtFR0zpwfSYwsS(zg?er>AaE4{c+64zB&RK4kL-@5ltd;CA4{FTjTc7<sRaMX
zI+e`Cue|Z<4TrG>{1=ZKapjLM(|zhvKkdM=TR)R^^1R3>wNW)T3wtQBr)X74zI!V<
zd+RxN4bf&wBuZfM?2G-^eH~9xm;eq;;;sx~_n%_$0f}l=|7orH-9XPP1j=cV5;X?T
z@i)0CCC)h|KBLgc!lK<k*Z+xC6-N}(^&JO3rxH#B?!^z{&nG-7e{fP$tM1)}Km5uI
z=Kie%_?vNR>R}X0S7_A@|FJl?YiDJDok@xwN-CU84nX6*1YKq}IwVjd&_V^5^Ys?_
z4R4nZG*lRN3^LMZ#0$6uS5Hf{3EAXzo*+^fg#R&%hSPw|qIG8QTf;wOZL`fUV(JE4
z%ClA;o>0i?@8{HK@1H&C1Ln9CEa0*50i@S3E>Kcsc+3PssBt@F*JRv2DLR{&d;V|m
zICW!w+p<xrK~N`-ou<RROeexJun<I8Ml5X;!ZI+{$CdoM&CiGoG~Q30A2#tzNDjhZ
z(3s~TE?>WQOyWc<M%LUpO%-nEEXKJKo`QqE^!4p*iluF>h+8YA1k>B?sfhsg56Z|5
zQNU5IE&P!i=1=L4C(1pcGXC(f$<Kwhj?tM??6n{H$K*Urj5If2!07?PQX2044(v0~
zK9DScnMB~@<;-zxd~B55ESx7m1coeoJyj@sQledBhcylf;vX3S@J9y2_hCWUQfI@Y
zf&fYgCXGh`56OUcY&3lo6RHhf#MVbzt8J6mGUEaz$l2X+dV(6pUWtT8=5R|P7HT_b
z18p2PST+z?d>n$xr;W4YO@67*(z*8&x_MH1^`b?f`0LlLu<Als@k<2^=`U(6svvU;
zwt@a8LRO&bLg3{#99Lqap_8H}_k@Oi8G`33CkZnNPy|A!-qRSl3C*$9ldm?Lf!oYi
zexi&ZFKa8#L=aEzciCLF_d~B8T6}Du4Q~frP^;l`O)};wG*CWMsqkHV+Ep=<*m)dp
z2a;c9;w6S!hwNV|XbeE6X}^8exPd+L<K2!{`G@e4tBOK%3>NKh-*#|kRSZ#pc^JMm
z@F|4uJlrenx8W+U36#<u4h2Sz?{k+Cz5vuP8MCf+%5ZN*HJR`|9oCko^j<Y@)hEWD
zXf6Aax}<vshc|^WK3(>Tk<E{(t#9Z?(S-c?5MvZlEugT<ICCIG<)kNv?0379js@|H
zV%F3V{WkjeGs&n*04%HO5Pq;^zS~CsI1A^lBX~bH!APg`pooE4v}`?>^zCGozP~Hp
zXZFfZhhvW>jJSki2w1au+Xw-p+H9^jo(|DsJ+GH{s9aoBhe-Z={S1PU{czi$DV#8&
zKezyh2QPQStUO5av==keU?Qi*)~;HQ**kLKX2N)eFMPOVM%E_hWA0#iq^70m69d3+
z;Y{sUmN=`(Ii1kXVL-tA^t*>91#C%Ehq*L4$7t0<n1W^^1~t0?((EHER`Rd-)$*}A
z^1+34kaGsQtB_*A>wH6*z)wkN$Dn$jakpv#<nq=jjP&%^A!P4*b#+Gdei@?$vgn@P
z;ouAE#yl88YDL+d8voiig$4|`4@XjsSe3d{oJxWUTA}MXa=uT8II};S;u5elvWK!k
zI0Pnpg)eVGN=4Te6R4>G05J|QW>|`WC?8*%RN0~Jlf?PQ`igbVm|7+WW?kEbvpe9L
z%Iqj4PXS05;~)vHsTp*SDV{?tb=LE0I&1cPS8M>NVGdt}kRukBzv{c)5uLGS$s6sh
z)>&FSvFT5zK0;`EzD_%8#@^mCdbT9tCCMr{AGx6>!z7sOZigo9KLKXugCW^lzjlwO
zM;L;YQ<J;nkv_@$)bFw(^2~&UOY^RJl6;z9R-6A*egg$A2PYiJmVd${l43B`cB`oc
zler%Soe03h-~HXn`9IAn1Z7NZ$5XB6<km_dT!SSP!Di1pK@CqfKm-nmvvqh<vsX&q
zJn5Gv{}dc%^NdGzk|mX|^WPU+xZET6DaWCIBMGtSD<+v4{HYksj}+yPEFcHC*lb9@
zYMY@~nsk5BW_=fzFu17DrBEiIRey43j<<K%HaKh__+Cd;V>MTuR;i|R>%#`)<u*OJ
zdH%q=7q)x)$(ukg@Z{W#IoG6>v>{NH2`m&E_-k?!Ism%x58zo=Ur!l&rYu!tm$3Ay
zSVh^EnhO_G`YZlt5zdmcwX&>c4LLPV4kg!x6jPx|0LMgvOd>)4<I!t=BkrwscH|I`
z4izr*OvkaCAJ7#hbK?8<B))I-Z_jwoW)LF!QPM@aC$@cZLCz9?-eqvlG2<!uZE`!}
z{%IgDIcYR)e7I*+;LC7v)7_m5+1ZVEi>~6}a~0HBzXC3wgZE|T|M{!P2S$ZW%`~P5
zEwOKE<z%S)KPI~LiIwx8UPfwZO>mVHV{$2_M<TNo*Zmgak?jIC1Nt>WE9*Sl{+m|5
zq3?sSoKa=F7?mJVo%e16r7q&VC_TA>*YBNB8;ik>0@&e4MKRl?K==YVk(#zDNn&Q3
zLxa2s5gJg^5Sgi(bIO1XJ84|t8@iSf=ldP?76Uu5Sy0{0B!9sc@TDqPHX0GICJzO0
z)JkT3!ztlZ^@EuR0sjlLS8S?qmpw*+od4GDuBnzA($DqEAfP()CuCd1ddlxN5pi^=
ziE-{s==T=WfJM=-<Hhh_a+F?K)P9?yW#MNipt)Vo+bD<9=gy)=XK0Bg2Tbv909isT
zKG`YAx{WhLx)T&!vA!26-|iwBIUXHJtu``@j#05wKDa+00i`0AHmS(lZG%MFV=boI
zB6g<HKITgv2CF1;VrS^VNTY^{R#dVp1gY08Es^FgLn+@B%{7B94k4miTjtyvJC(a@
z%`Hb9*Hyt`hH`5DWZ%q0=8(VDd=1=(%8#-HzG16B#K=)*vfd$)<oF-Bot`I$m!we#
z&F86zhg&45fo9Tt2Tv+YzdKnJUV9=530iH`{p<*5kPQ)vvE~;3wqb$lU)qYOgHJ+y
z4_hQN)6?_&-eG&j>ktAjlWRr`<N5Lqh^%0lx%_2&rMspF$LBW@dyg#R>7I)f4ZZxL
z)EddIRjXH57yGWHbUb_De#X`<#8QR-ga6;JSa}4#8$?||>-_Yg^DwbQgH!*NiXs_R
zwpm-kcr)4dh*N`&JAl8^Pe)|G@g_{MZP=L09LU~}@)B<dT6MzQgZ+LPF=nm3KtuhB
zEQw&Bj{E!py^-z#mJTz4RjwHuUC1G)oB_Mxm3dm~D9T2(S+r9hUtizs>1XTyOTPd0
zj>qOL;3mCRO*5IrDcNzl`yDr}^W{89cT#y$sxKjbtKSh9_CSWd+&1XHZ7e4}CF609
zWVh5xR@Ud3c{_~zD4hLSu3UsWjxG=UIJ49yk4a;&XH`s6;+#Zg52)c`c2jl?O!urh
zq#$<qo>w>~?)WZR9Ak2PQp5^JPPav<Y{QjD(j4sC^jB=%!MEu=iz2Q{7*P`A+y^RK
z`-VD$B!Y1;7Us>WO1I8?cu12PouwsT-N#}6V$)bR<Us}pj;z_kbZ5!wL~zjb^WChA
z7*ZOn%jLid5iY<4>ZMMYlq3z3reKf6_fO%c&kIpjpiRUIw7w+3_>Ud#^Dys(E2ZdR
zxaQN!1hrI)Ch;$ard%muMK?T-+;~ZYBj^&BD8~gErr&(xWWl6Z`pJ4Y$*oQpQYN06
znT6StDw@eHk06YHl)=e4(<`&ghT|lA3CCqo5^7?U!Mq#zmbfsUtP`W^DkXiJXr?A;
z3@ZF!YVPyr=KQxLj>#XO9Jaq?&1Um?Q)Z}l3FHs_<fCKVNovM3kjI!#k^4Z<A|y$t
zFr6$hF`5Qn=aVnmIXuH-aL1kwW8$}-a`EuS{XO4;%E*|&n<Nv0J(9EFz^Z>KcT=4U
z#mM}-7U>@aV@iayZygaKU79sBHUdPe)x-5NS(N7&Mq1*-{akyDw60kB9>h-tHL<#B
z3mHjnA0J_45jypVNTz7Cp3nA-jkSH86hMsJShFs=_pH;nBhLyM8%n%-UVCZKfjb}J
zdG>nV;y%bV-Yod#ID1Ahgw^*j;eU1f9%PGp7Bea{pdf^n%i7jfll7I)<4G5fZHw|h
z3fq4|ULgVY8Fp&+MD<%T=_JNSMUj}4hjvnVJQw=iC^BG&hU=~-PvLv$8L2_!-?Yb*
z7#tZW`i6$jsrKTxnP-QXGTGZD*$?aX*@@yHzuA*w8%&l!kA^H?OOF?3xpzqp(<Y4O
zs3{(RPg9SY?wJYn__WQVbTX)h2Hh)CB!04iKt+O$oX&+%G6H$=_ALV)i_jyzBSfQ4
z<cK#o{o=V8Jc$7Xkp$x5rJbj6;>JFN{2?|&3<%9;o-cI2I9ub#54#3Ksi-oqcKqUO
z9zUagq>_6sh0;&H(%Lid{@^?sj;)9MS;x8mFiw|`7gL@zBe}u?N$e$~PWB4b5O88x
z@mga=ovl!ho58iirtC<LXY+juxlbv;F+qwcWiT3I+}gEe6Ef9BgXLqU?a~fvFu#aN
z=*Fx1>=`msPciA@GDb8a1U#p^9^{Oo+baWZ?ddGv)e`2%?b3_!@ua?=DYI=|htIw!
zyRVPu*pyEARX>G6$%@qxmy-y)N7=FU{_Fh=P_cn#P_(_><t@m{PCYyBqVqD!9r<h;
z+o}xYGBaRMuFO)#{H-Mmv^C(WHZQbtGx5ez)TqN?2_M@4KDMmifxZwR5B`9^wFc%L
z(>D7_!|Qwi;TOMCDyp2sAQ6GuceoApx@%gL_x14b6A-BU2NQ6U%P{iJf!nXHF0L#`
z#1zdHXT8Rko^0CW5dVtu&kBkhtXoTN&^0Af9NLZqNI|D?M9yR0s<DP0dTh)RPv3f_
zy8tYE%ao5tZ$aJiKX)&=Y6UVthC@Y%wb+2q_({43kfwl*Js>`ECO9z5Mc{;*K?{zg
zjURHLa!KCyJJ^e|;m9qU=W>#g>7hH!d@BIXKrMnu`1+jJCDk}3aKp-8oT{u6xPNfS
z3H{Vw<d1>kjY};bc)qQZ<5<EaIXJ8E)vy&G&q=VW@O?8H+aO_lf25%7k&UeC^}arr
zFPk3zaT~2Hg;mJrnS$enXrTT<|KeB1?M#V7*^~={GOe;5;ao60r-E69=+1EwZDKaS
z-L<QJ!2-D}iAe{ui!Oe@+O%Jm6Fh{Mj8X7mm(|?c(iQ4AF?&j0GcaQ(H--{aLQ_E*
zx$<+j#4AW)0JC=lN13}leE86--8Dle>02`|`m%M<I(R*gBn%utBRH}kr!zcD0ipM{
z^*h*%PcwmFtI3<BQ?&Q3#rcLs??+;b`6VPQNR<H|q2`2$9m3!ZXO&gzecoU;|3t|Q
z^?I4Sp6mI?Yo+58)Oq$OF%fNw%0@jK8=I$lEytc(yz?Ian*T0OMoDS=&NE+Fa8@nF
z5FqL0R98}_J`TSnHWgDB7dC+qZ?OhW<DYC3l90&bPl>gM^g=bA`i$w$<h!ZD__z7y
z#rMC7(J54@X9yiH)niROm4xY|UvX>0eD!d*KWvhJxJhstJwFDxcJ!qNf3S;ii;^8-
zvLpMg5BawOd+_!S^Ut4-aw^#Ly61B2WeTbJG-gnmtR2Ec&z}IaM&g1{Qvel2_eW%7
z^%6Ar*DFcURTbBUy63=4N&w+AJh(_31fV2?KP?=p1_F=VCt^OI+Ak%&i0TEJtQV<V
zRF@z*J9WR;OHx#P4#&8YoC%!+{!ji~k8;g&mb64~2~{RHE(z-2R4n=b88bnU`+w(Z
zNx_0gX;PO3c`o8i5lZuU!%fu6qVeIAJZH`vA){1ya!_-kx$6tPEgZQV8c$D#cy!Oj
zU3A$t%4C%*v7~bC9j1>rzP`xKTm54$V;m<SKF8GHoQ&;O-<u7t^v2e6ksr%Z4?BH0
zq9<$Ksk3|d>=ttaP~%b|wQU87i-$&Iif@>pf2l=bS3ol+_Um7nppyb{>K4tWenEG4
z?!Yc@0`I33lZ4?tp~+4(SuiN}TLL!VxihS)W;F9s-7ElQJZ71d1rta7&|E3GW2|F7
znqE|`N<)yU4aagCe0Hr-?6-mHfhh8Y##En9y}v<Z2u-{~v|J=@4NNJ^7lP-#&OE)$
zeX`RaG_F@VbPT{%I2VW}&A|X<hz=Jt7HPLBC9nB|M6~pR!Z>#)n}cV9WIg!megeWw
zuwSz5w0=oa;Qp;W;Z)(|GrXycXhs@}JH}ml#>TRPglfhvW~26;i;HU;w8szU(8#Vf
z2=geUemI_z;gP0qu$}e@lG4;toqBKBW=W=Lrq(<IM4Jkjz(wL@q**jv1+aCTAr70@
zC&z)C;3QaR`Ga?a3R&tGtXCqPvgn^>;7*AjoUkBtp?*^f;MPDzCve$7iv2s;&s(;;
z<m(QK)s><{S@qK;QuT1sAd>e$e9obPsU=@UZM2HjK7#N;;Lc%3?-wpBH{@7LJZr`_
zcQ{B866}=#G6X90r4>$u>|Qq4x%%lU!>jfqFoTf~B)qgMOoEE~sCQ6vvwhHwM2PP!
zq?n_72rk0W3&<AyYfnUcAu^Ws;Ao+pTsVXXke{7}4h8Rb{?o3}EY*6#i&Fl@sgCbX
zD0A3OoB2=|555D;syHFVXHC)@vK+r;#dJD8|C)D@^HA@N@(u&YJKt7SRi(}1a?N;l
z0`A`K=8XDo7&fzql%Vy_X4{yA0)F5w1lZ7pA<~+56yrjW^7s*aVXZ(Bb0k-__~PV4
za2ZDoT3lLSpPqp>Bh^uGfmKLhg>E#EJQ9rM!^M-tGc`_Mh42dY5HAHp8zV_#kFIXh
z9iCumZ2uXbh5A8p;Lc{ee5Q(ZsFX0xt03}>F&A74|Cs*#*UnR4mVN#F{gk=IRNzSE
ze7!nGU#OqPj(lg&`}$--=V@|=*-wjY-xa<KfksDLC%LK;QZ3i6wo)qnOE#vQx$U*o
z;!_z0NKZk7(G|wnmt@QwQR7Jp!k=8@_ULh(xr0-v*P<_hlS-$2Df#FRUmWfq985T!
z_^hka;ilwz3HIzQOE1MCk&8;9Bs$#wmKVUUvx18b+h|~P0`|J|>qGdROES}HO|f*0
z5AUMPx?bnVMpEy_EvAfpJbRVx{K&5tijS8!D8!Sq84EfWkAGcUc~&G;o}92egN8Xc
zk=7GsRYPlUtgqBO>&W6ajg5`UgIA{CQmyI3qdYE|VnNMH(|_hdG`B!9a8TW|{kh|z
zxWJ$FO9emnW94>DfZTdAa^{(Of(YjTCrYB+$#a&nI5jpeB1QTk@<GrJy{8NWxy?Pq
zuUi(=)5;-ctU>nOM#~W;3h=Pk(QC4syiu)VoOkt+{R)FSlu}VMS<UP9f68`#*E$|E
zxhWDFetqNXnm6iX&z>)BZ}qk*R7uh_J^z}>%l_Qfbq{=(Mo*%gCirzdfradKvYP{>
z>G@|gF7czPG4YX+TF{EI8Ozv7DjskW__#vxhveTTB2fj!Ml2?@FACDL+xK1f4X&o*
z=OOZ41aC)|`Br(AqLGnYLSf@gQ?Y3n%VHi5ZLsRa)F{L+{tm^Mx`mTg7z$V}Y+>Q{
z&b!!1O_@nmJO0qOLZ?0iv%5yhzl0}h3Z3bOJh7CeRlQif$?HkLYU*tnd+Z(l+jx}7
zg{HK2lK}j0h4#PvJ;f6H?Z&oYK!T<$E=z-@`g@%!&}SszfqF^*=TIl-kLGAt+gP6e
zAwM|$Ww64j8v<X<Nf9)*eH*(mKXC!ISqWKxx<aPKA^+0#fBO76eP<x+1<$}2gQepv
z5Eg@!)=1Q0pLzqNo?Zrn1NiLf$wUUk*Cek)2T0vHkA?caf6)JbP8di;_OAW==bQl*
zkBk#WbN{!Mm4>7feIT)R1-e~v7>xfzl+3fBTJhNb!n1$VfRk<DznX~eTN^b^j8ea3
zP2|icjc7)02J9p%82zJ1cR6DtfJ#NvnT2AE6I{e8LutG(A(`FXH#RQw-UlMJ*)~UT
zn*+6YV>fsC0e{6wN4jP(zA@m-I*e$?XgE*5wx1nqCp2aIuOJ98#yCkX{{C6C;fQ?#
zR3aKt>`mCj8X2jsBGW+3$mGw@m^=l*U}>8>`~bfI*kP7YgoqQgEII;g&dN=Wq)E9)
z;c10Tp6;%R0E^C613nE)!?Q&8BnrV6DoURXyGCR|A<pRI5W;kixa`issi+*V&MbwQ
z#;?a9rU+7UBGL&XGh>UZc43-}r?!}P;go8%-k&Y5$7L{wR^`96d#kt~jm%~r)xD80
z^PicxejiFUPFmY=wA;WIvue0_f(ckg$OHz-G!xGd0Z)b$Bo~_*jT*J%<UJ<z3>(9i
zd0zPxC(FS=Q}mNO8P3>)OEhNZ{NW@%bNV?s-|*M}Yz>=&zDuL2m`=kaZleyPm=f*l
zyVIr9-V8E$vfYNScgz0odpw_Yk~G7wUr$M;1lx&WgAXqo(D7v3rm7v)Psq&nzQdQ~
z0E<YT)YBnUPbhDS4c(sVf$`8@wKPwo(w~<C8kD$kQVEkgI-4wZ<o56#=vlWuxyJZn
zoiny}lG6oF{Dw4!3p7_$fJO6WR=H32(%}s$o{eJ{I1dPfEPA9_dR0hoecgQhnsDhb
za3&DLAR@Rj!WqXX4Bn{jdJZDuk}Qe4%kbE|qK?ZUQj+~P|F&&<b#JoCApTciqo_>u
zqOQ(PWk&0y-`>;Mox!rX)ooTC3_S4OJYk>&aKEYp+1ifyOXm&d^~_QfUM=%lgW+j7
z=)Re>&YN5`q0=81x@H(0<S?jXd=0}hUMk)<pz!uj)*&Z0v~7<xdjfxCmUqmg<-(sE
z@*pUk(Z<^PC<XF%Q{aW#5zt&j<%MRB-Tp~+7n#u)ifwOhuC2qHhi1V~`nK^rBg?k2
zn$%;MZ(uQJN5Epmh(0K3Y}uvf6}RWAha@QXh(MK|T)ivAj@}FhBa)PRt=?Qn!#Nm~
zHuZwE0bbXhCEHCLEXisck>Ivje|OSJD)>)6PhFDi`=aDr4#X-Vue8K3ate;=a5I|K
zws|)<H^(0?fQ4!&7ooe8-g6lhh1Z+jk59QYoq;4#amCj?VXo^rgGBOHNVII$CF~Qa
z`S#7KQlAkI7rHB`x$(t(`qrL^D-PR^g3SjQz=!B8(Z1h|7yI?pn)%NpgNZ|MB3FPO
z9qCmY_tB30tK8S+_D7RQW0GzlhJ)3YZP9wuFm^7MIvoC`-#W^ifp9cvd^TMSI;4%A
z5QIDnpDFwCMUt~o2ra)X;D?)<;}W6=NlChlv?8k?F(e#s1&Ou_Wt#YFH6NMT1bv*2
zdR@S<mEvOE1g-D5V}M#Y{DHo5`~PAE2rV0X*z~+KspkI!Pfek)KRYq9?f=polOG_v
zJ%M6mxBnGQ^=B1^L-bRR$fE>tm1THCz=IR;4)F8T`$yA!JZ@!UaNrn?h)LOFuSd6(
zm{88rNO$s<f62Ed|0Q{)hXPVcObHi^W>ZGluOBW_m5Kz+{=cy8{%b|~l|BeLn1vm*
zSkn^a7`WmMsZ?nCsSs%xsE{mtOpn3%)3>&wTyEYe(tI(n=a^i^gnaGjpm~-5_65Iu
zGj=C2`of==YbFI(lCyuOR{bLHe=!>W<@BK6S9fIf40NriMl{+4AXR6f*O05{q#?1C
zgwPL$O?pHpJfmATNgDIvsYl;Spho*?V&2UEX$Lijd1n3RuP^l{WxiUC@-((fv1y^8
zjX|Y0I7XIjK%U*t2bxDBy9>MeRmG~W2$y_SPv^IFVrQQ3h7&z-!pkjjI=4NduH8og
z=c+j=?syF?wonDG5j^5&j9SDF-m$#{QA|fiM|=vx9U7*g)t^pVd{o4vCdPxY5$D;$
zGA=UhW2y&U5GK#MEigRb61Q9jek7+#4hCkfuM10;QOu}b$$G+$r*5T)gOM&}O94oK
z@ble|A40IJ;?yiRKflAc1dJDxNi}!`z<|>WO9@%pW;HrqpL4e9ncv{z_oe%P4g{YY
zor6-hO3*WTopmOwY$#{Z=(J6<+7=7xb=W%Yv6Q*Ap?rP1%&~V|T(@$MU+9sjv7i^b
zR|>Ib-?~D#!2_BN^p9%nM7R64U*A1PZsbOK$&OqLH8r)8e2n%EWoZelj1LH43=Ckk
zI!p457543V*1w>d+vt+32MlLE^78WEk0lF!zF*)Xm!)+FzM|kdIXgZ4V1;?6H_r-g
z1(6H)jT~)vUU9IqyJeMSx-g{gm~c8>PHSs{Z$(%CQe|!rkB4W+lgG!Ko6ijl4tlo_
z4h|}r+rAkc1#Knz27$Co!yKKQZfhnTAMq|)wDV<7^VnDz0_>NKHFouXpDS3${_;ZC
z!kq57O5^W8e*9ROGn4oI$jD9OjFrwq-<}USh1t$FD;xN1y#{o><Z5vz^&i6*7q3=c
zceX}u@#4kViKlAs7Ws^des(V^I?zg&`W*$s?bglJMUPk-dAH2CdH(r0N4nH5NozYh
zqt`iYN}O~LvPw=>a>pt(rs$*uUvCMC3Eh1pT}))P^faEQal)c4Tf<9r?S0=TJ;=_E
z)J(d6cgeiHtTU{}v{I_0bW-+bXco`CqbTqCNjS8xJVs$>&BrXocg<45(}G3YrW>9h
zw`#ij$BquqExf$mLy9!X(vBr(&YXFXyOb7;d+=KBeNNG}>gjKXhJ?k$#XDU?E^HLu
zKcChZ6iRnntM4dnIWMzjLgnQWt2h;h=XvKdTaZl?iSF#R!%UlnOE#CBT<buW{A0_u
zSMyGNykKgA11c{|T~?HTv|b@=S8Z)r=;GU2SrN%v$~BwMj*n*2MSgzwAw9PxeZ<D@
ztes5(?op`w)6h`coTU9LwN7ij{aEvO*g)Z{?d~HpPwviex+$<&(9DPRv%6^b!n(-r
z+KydbJ#$|@dG6b{!{w`Zuf-l#HnuzRU#*k;KWH+~Y2?yrP#S9$w$9U0h_TF4wfbOn
zL8YwWZhn4`mXciZi0A#eZ$IuXT;qw)S2Ro2&#l34@@BJ2JCFSEd)1cKyK1X$zDu-b
zlKc6Lr(-zR-OXCO{_*2!4Gj$)hI@N@dOE&*Dao2C@HzRyg&FA0-mX&-^>VXH6LA^t
z7Q~v|`f)ELyWaRsI=5$<(X+6yC}JI_=;n9kTjsDYyO*jqkb}=LCWdPywv(=?yyVvA
zy0Gz4DD4YOdY+kxNJvzlIj0q;byp-gC1v~BTk>NLBcr2vWw)^SW`%~azA)S$L?0P`
zy8GIQp?wW&S}1pL(f5MYXAK@=Rhs`C=nmOt_1;)5IXM}Rlf1<JE*86Rp(8I2bAC(T
zZglDXjB`vh@@Lm>k<hw1)VFu%&s3(qq0L%*WjgiUhPl6fZ0_xezni~j%`ECW#U<0Z
zsc&4*FOi-`eaH0IuKCnAx7PfB`&Dno&GgslEm!vE_}SaryKU%UzNcKihkVR2Q5+nb
zlFs$~`>?aJtWus%3>G?0c)A^^X@3M{S#!_HdKPbZJLp40LzBMScikvHZnC;>qx;jf
z+|jXRM$6A$ibPovEsZKKKQwyN*=*+ST}i5!Ji_vyhJ>sFPbLuZLO{)rW1gFybhmBD
z#;dPn>zC7A4CL^%T$u1_a9Aa!x^QsREp2@Yueh(|r6$%(t=C?(_C7iNExT>!i5)AI
zaKR$8X?6=*gzh{(dTVt|<(Z-fbvV2pqQl|2Qk_o|2VdGsSR7f~Wp|R%tnt7~?|qPm
z8||>p*;3zUlVp{AKPxn`do?R-QbobtqVruVU-qr!99Co)JfL0ums8W7pS~t@lt*2{
zMN7-Mym^h-9r_*zCVca3y|y#7<Sgg4v?Ru(HCLY)?TCaQP;Eg_<4U?~H;O?pu@q$M
zDak4=ci+K6=iT*~xrRd~lGh-zeL>^k=(0ugGuQhZQ<nAX8Q%HzL)S?m7UkvcKOfAw
zaE|W$ldZRfe7qelh^!84UlG1G<;a%4tFmq@y=V6iJLBGNniD{yg*-F6Xc{~`IwtGv
zEgQW@miy;m)7k#}$M3D+jyBpEx-#^|kdu7BgY)S*X|r<_o(Mh(nsq3}ro2J`()mxH
zJn(btWxj=XuS>=O0g2W#T<fiBSFfbISkW`sKkzO0#hpODpRrM89=H^Y@(KzHb}V0b
z`62UQUS0RVi&y1!an1pS&mJgVRjx1<IEh2LaypV$(OqPkdCqd}@I!^L_Kuvz<Fjzl
z*$rP_Y+BJY$SuQwD_>B`ztUT6(cSwtueZqfE5?*Z;6S;r_T?RcxVL8)pWo%XtM}~S
zz`)Y`D+*Fw!j-+SW`uk)9oO)F`1o;O^}Hn2qPZnq{X5soL>3+P=h%5_{`p6-f={+u
z&sG`{e`wyFrkGOQ0m>$!U@q-S0qDNDSa|XZ3NqYMpz^*7h?PclSjVqch&hBcx8wT3
z%L808{(_Es;~LCG_A@rd7aHurHT!eemkWxb$Ylzw<PRxcTGyKU<4zopFV-w)fajNE
zAMK0zDvhstNDponE#01+T4p4<dx`1yY;{LTS#GJ!%*=i1HAXuwzn7%TFVx%VXjH={
zz{&0vDZT%um541<r)=NapZ4u_{5u7%AB=4<=N%E-<mR!?Q256B@_QaeYqc-=N~u~k
zm{->%=jJs36lFi1uD~sYp@WA@W9rc8U{n7Uqvh`3Ve)z_F+LLId$CFPdr1ixy(dF+
z&p(pxJ|H4Eur+DK^!h*$zvLH((A)!f51&4Mbcd>3er-V1WsTu0s^+Fv6_PFKvdLOe
z8{RT6YT9P)Ty1NZ7|5qg6WkrNEb@7eU2SUL;ESH3(+?k{9Z{9xj{dH#GL$*wWd4~~
zSK?bv^Lu{po?$GCMyCsNXgy8VJLSygJYZe#VBY@0Tl;z>v-?{6u>DF*N=s&{B&lMB
zbXCi9G8*q#i6xS(mQ}uK|0AvLs64X}?Vg-xmN#o>=bT#gSYxhj=ejeohT;2;qny+_
z?=CbS&I*6t(b6|`)1PA--(C;tOx{Z4!zaYP7Rxsl$j3wtawV0%NI5MQTor!0Ijy;K
ze|@`ey_k7;o4rDa=9>JyIZbPZXG57hXQR}q{704fEvNn4`R!{2GlzaQAH2T7OF;a_
z1KTzILwzUaKUG||bhxE?!R0gW<egu!FRp1U!GY=Z&3mehGpyGt$|^-G%j7*u+qZz@
z`W~i&)GxV3-PSqU&gW+BJ}UaL=ZXBd|1v#1>MOmiza(|_1u9!ZjdrDkW7B<Imm#;h
z?)#~}TFnnORvGS#@++gg>|4A4^V<@+sIpr+DKWgygnat1be$5~p}a>Vl&-R?-;K_$
zuFDjU$i3Rz*fd*n6#V~k$O}t;)FM>X;ib{GPu66nr*!bPWV_F@7YDERp83%E!D8o?
z><16t3_op)E;IVXS77MrtIZ~uAC{k)+is$9<+P(?Liichl#wHv>s~j7ny&f$qNp#|
zZPy~&`{?yD{+e8>36<}n`1nG3WEdJEf~QGkXhwHf%RcN`6{_=)zOLKW`@T>8#;k6a
z{1l1P*sbDJ-i~_8#{_7`+{t*XVUS&Ubn6>uR^f|M11IffmYQGs9C)hnPJ?@4y46F!
z5@?bKS1z3WbGWs~U*g!-<;$Jx(#7j{jtYm|EN{KB=Zfu^$IBYE5+}d<fR5A2E*CzR
zv#_!Xmx>LZdwYkE$K;Y)UT@)wz5WXWg$|f+U4BK`%g5Ju!PyslCNTrOBNrj9*ni@|
zQAT?2f@4vqc7(_|&pEzAW2mOSIdH8=@x}7n0)CoCF>ID$(FY&5$crtbWv!N!*d)B-
z5rkRTL2wUk{KHR3Ko1>6m-_9@W52PqoOxKgG390BksBgQ!zH@f%f)%S-@VM@a}t#_
zc(>xpI_qSm%eg55^}$VIXP4#|3_6Q>Wq4Yx>qE-Q7F>q5U!gOq<_G)6CmvM-sEaEW
ztuO1_a9um-$I6dAX5!(i7A?u=EVX|bzi@w4*{h+-`y0Gq*^qAE-50y2{>9wYv_~d>
zW`a%cT3u#j7GIT>YSNPRm(1t#VvkDEk5P*0U+`hr<K?y0pIeMC^xdsrBY(i!ucO6w
zWu8Iis8hM(9gp7AC!KkNs?KbDd*Vo@L^X8z;+8A9iX6utC_Hi&xK3BP;t&?}oWYv{
zr9ydSoWsv}LsfZ?E%~z8dS|FxNSbJ8rgz3Ffi&aQt%2eXL*5UI7q|Dh%|CirqbEZo
zR>LpvT~Jpt|C!lsp=?3?BiYG`rFaPUqc!NYJNU4NSKH7cl-Vfli*oIm1s*S#&(lk8
zuF6d*N;7?_rHRdFpoI}<LuDVbI2-&;xO%YdXNVU4mDrlp`h<BOheRX8+s=xz@$T4D
zRI;}&)iU&G6Zfkeg#i<6w&z=M_~@z4>#j_9twxFZxLxMvK<501G8;vP_?9$>s3t$$
z8@5U=VL;gHO1Im=>Z17@4IH#p@6Vkxi^rn=$p^>wKKoAZr|E-9b>T(zAD7Kj){fj@
zMQ407ENrF5@tgaELW-4#_do3EdSAq^|GaJUX2$)E%U?RkvE}kwvaDJc8LfQM#Lzxz
zm)D`}&m~#8()Q_YN#$9E4m}$;Wq<ukPGnEp`oOGRF9VF`xQB3T+CAK-RobQ$r((2q
z`KO%i^&4^e1?Q%-8jTvAgTrx~?mf_H&sHygX)k!cT65k@bFH?9y@fi7k2!-^oN#n>
z4CI(6rIfxkMnQJd0WHhA4oCYEwZ=7Gx+c|&?9Ej^i0Mjq&2X)L!75TM8)+qPrgZZ~
zL0{DJIqJFEwP&vMoer|hvk0(bx^U==p@Sw%K}q<rwGQ$!_uB2UHbh@G`yhN=t?AM-
zciG{~)Xg|j^|3xYwEdb^J8##Eth_PWplA8dhoTz;_xX&}4m(>~>b+j{y6;Bj@uUiI
zoyuYzn|*g2D!XjfbzUv4oh7!Zt+bIYMx|HzmQ@K|_zy$Pq^0E2o3L7qGc+CQogUXe
zO$`a^w;!-9Illdb)Unc`mYhM+vnH$FmMjWhY%dxre-RA{x|^tXw`bYg>juBjvA;5?
zXJKK{X0@(x-YrRqb{co|(gAI8wItPk#hp!JTG3jCu9tncZN8>wE4+?<9cys!+X%^_
z(hICCwSDaz!nN0owk=wV67^!ExRdF(;UoM_aDoO1T`iH9_%b-W@8cfqUpB&-^E7_U
zUU7DItJ?+xZR414ZAm8G17-mUpJDH*P}w~stkG8^Ogo)g?;=vEkXA35)sV<dZ~xP@
z>w7!Lu;_=5(%SaWddn~Odbj!PTPHT%aB$XxB!vTAMVE#d`7->6nON4ep;>VHqVugi
zxJtge%Qol5)-`@v?`JMME`0DMGamgneO5o5Pd*OJV;wV%Z2xq<{reI>`mEzgbp=uD
z^_H)6Q}I#$dVk1X$I<1;$-O7TPpsxrD@jJ*xMI`0Q)6BC_q|OkS){<!Mkk!|?AYSf
zG2U9yd0TpvugP-fExw#zo@f=YEG9S0XPueH%}-g=%>_b!n%{VRc1vWX@rr^+35u^?
zgnSxu#F<4qLgo9XuSmPu%g?^QP$>Au{;+R1Q;ai=6LX){XO@fD-#!?$xU|no&Bf~6
zViz7cPR5Z9ytG_2WS8PA<CRX>hn;wLw8ib7-R_H`<%}QuhMpMBI}!2nXG*1UM^h4e
zWQDk&Nq+ZWTWG8Kz5PC)UD!)gxPI1JyjK#fjcIn9A#`o-`Ca2KOV4B<Nea5xUAmIL
z*{$5%-2Ad|_kc(EW`%IG%CVd19L8L}4IKX57d#M>k}ENyJJvxz#v`oea9FGH!_eJf
zdV8H?r{z}Hez1^i>&;&+;WA>Es&GMVTS)PDFCkj}QupR37Q1sfbn0vwFKN<fkG>RU
z&l_Jk{4MBEy=_BX!ocHt!BF|kpsG1X9lKY5fV9x<<C|&tu@<F8*JnRS%F_EfW@=q2
zZs#Z@M~hPWsw}krO@BRG{fm5O_t{Hmv~P;C9pPn1(JNm&u^?+r#^Om3c6;Fe5|{no
cL%g47SE;|hHmFltO0G#pN?tN~&(ZV$4`HKUN&o-=

literal 0
HcmV?d00001

diff --git a/requirements/requirements.txt b/requirements/requirements.txt
index d093b8c3..d243e263 100644
--- a/requirements/requirements.txt
+++ b/requirements/requirements.txt
@@ -10,3 +10,4 @@ tiktoken
 datasets
 threadpoolctl
 shortuuid
+qwen_omni_utils
diff --git a/tools/run.py b/tools/run.py
index dca463ea..0ff5dfc3 100644
--- a/tools/run.py
+++ b/tools/run.py
@@ -92,6 +92,7 @@ def multi_nodes_run(config):
         cache_dir=model_config.cache_dir,
         deploy_backend=global_config.deploy_backend,
         using_multi_nodes=True,
+        use_audio_in_video=dataset_config.use_audio_in_video,
     )
 
     # Step 5: Prepare data (optional custom dataloader)
@@ -105,6 +106,7 @@ def multi_nodes_run(config):
             num_samples=dataset_config.num_samples,
             shuffle=dataset_config.shuffle,
             inference_settings=dataset_config.inference_settings,
+            use_audio_in_video=dataset_config.use_audio_in_video,
         )
 
     # Step 6: Initialize compressor
@@ -149,6 +151,7 @@ def run(config):
         use_cache=model_config.use_cache,
         cache_dir=model_config.cache_dir,
         deploy_backend=global_config.deploy_backend,
+        use_audio_in_video=dataset_config.use_audio_in_video,
     )
 
     # Step 4: Prepare data (optional custom dataloader)
@@ -162,6 +165,7 @@ def run(config):
             num_samples=dataset_config.num_samples,
             shuffle=dataset_config.shuffle,
             inference_settings=dataset_config.inference_settings,
+            use_audio_in_video=dataset_config.use_audio_in_video,
             model_name=model_config.name,
         )
 

From 51d5045cbc396a8f284a68af8c3d2f4f71f2445b Mon Sep 17 00:00:00 2001
From: ali-12388 <1940747290@qq.com>
Date: Mon, 3 Nov 2025 19:45:31 +0800
Subject: [PATCH 2/6] fix AbsmaxPertensorObserver scales ValueError

---
 angelslim/compressor/quant/ptq.py | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/angelslim/compressor/quant/ptq.py b/angelslim/compressor/quant/ptq.py
index 3fe1d325..8aabb1f2 100644
--- a/angelslim/compressor/quant/ptq.py
+++ b/angelslim/compressor/quant/ptq.py
@@ -14,6 +14,7 @@
 
 import json
 import os
+import warnings
 
 import torch
 from safetensors.torch import load_file
@@ -193,9 +194,19 @@ def _convert(self):
                 )
                 is not None
             ):
-                self.quant_model.act_scales_dict[name] = self.ptq_hook.observer_dict[
-                    sub_layer
-                ].act_observer.scales()
+                try:
+                    self.quant_model.act_scales_dict[name] = (
+                        self.ptq_hook.observer_dict[sub_layer].act_observer.scales()
+                    )
+                except ValueError:
+                    self.quant_model.act_scales_dict[name] = torch.tensor(
+                        1.0, device=torch.cuda.current_device()
+                    )
+                    warnings.warn(
+                        f"Not calibrated for {name}. Using default act scale 1.0.",
+                        RuntimeWarning,
+                        stacklevel=2,
+                    )
             if (
                 getattr(  # noqa: B009
                     self.ptq_hook.observer_dict[sub_layer], "kv_cache_observer"

From a4e74c2683f94e0faa2dc1fc144c46b2cfb91d37 Mon Sep 17 00:00:00 2001
From: ali-12388 <1940747290@qq.com>
Date: Mon, 3 Nov 2025 21:02:41 +0800
Subject: [PATCH 3/6] add multimodal related lazy imports

---
 angelslim/data/omni_dataset.py           | 4 ++--
 angelslim/utils/lazy_imports.py          | 5 +++--
 requirements/requirements.txt            | 1 -
 requirements/requirements_multimodal.txt | 2 ++
 requirements/requirements_vlm.txt        | 1 -
 setup.py                                 | 6 +++---
 6 files changed, 10 insertions(+), 9 deletions(-)
 create mode 100644 requirements/requirements_multimodal.txt
 delete mode 100644 requirements/requirements_vlm.txt

diff --git a/angelslim/data/omni_dataset.py b/angelslim/data/omni_dataset.py
index 6f2f6fd7..52f51ed0 100644
--- a/angelslim/data/omni_dataset.py
+++ b/angelslim/data/omni_dataset.py
@@ -17,9 +17,9 @@
 from pathlib import Path
 from typing import Dict, List, Union
 
-from qwen_omni_utils import process_mm_info
 from transformers import ProcessorMixin
 
+from ..utils.lazy_imports import qwen_omni_utils
 from .base_dataset import BaseDataset
 
 
@@ -110,7 +110,7 @@ def _process_and_append(self, messages: List[Dict]):
         text = self.processor.apply_chat_template(
             messages, tokenize=False, add_generation_prompt=True
         )
-        audios, images, videos = process_mm_info(
+        audios, images, videos = qwen_omni_utils.process_mm_info(
             messages, use_audio_in_video=self.use_audio_in_video
         )
 
diff --git a/angelslim/utils/lazy_imports.py b/angelslim/utils/lazy_imports.py
index 00eecd8a..a052a338 100644
--- a/angelslim/utils/lazy_imports.py
+++ b/angelslim/utils/lazy_imports.py
@@ -156,5 +156,6 @@ def __getattr__(self, name: str) -> Any:
 referencing = LazyModule("referencing", "speculative")
 
 
-# --- VLM related lazy imports ---
-qwen_vl_utils = LazyModule("qwen_vl_utils", "vlm")
+# --- multimodal related lazy imports ---
+qwen_vl_utils = LazyModule("qwen_vl_utils", "multimodal")
+qwen_omni_utils = LazyModule("qwen_omni_utils", "multimodal")
diff --git a/requirements/requirements.txt b/requirements/requirements.txt
index d243e263..d093b8c3 100644
--- a/requirements/requirements.txt
+++ b/requirements/requirements.txt
@@ -10,4 +10,3 @@ tiktoken
 datasets
 threadpoolctl
 shortuuid
-qwen_omni_utils
diff --git a/requirements/requirements_multimodal.txt b/requirements/requirements_multimodal.txt
new file mode 100644
index 00000000..0c073e6f
--- /dev/null
+++ b/requirements/requirements_multimodal.txt
@@ -0,0 +1,2 @@
+qwen_vl_utils==0.0.11
+qwen_omni_utils
diff --git a/requirements/requirements_vlm.txt b/requirements/requirements_vlm.txt
deleted file mode 100644
index 87d1012c..00000000
--- a/requirements/requirements_vlm.txt
+++ /dev/null
@@ -1 +0,0 @@
-qwen_vl_utils==0.0.11
\ No newline at end of file
diff --git a/setup.py b/setup.py
index 30a97710..bd22c046 100644
--- a/setup.py
+++ b/setup.py
@@ -51,14 +51,14 @@ def get_requirements(filename):
         "all": (
             get_requirements("requirements/requirements_speculative.txt")
             + get_requirements("requirements/requirements_diffusion.txt")
-            + get_requirements("requirements/requirements_vlm.txt")
+            + get_requirements("requirements/requirements_multimodal.txt")
         ),
         # Install speculative sampling functionality: pip install angelslim[speculative]
         "speculative": get_requirements("requirements/requirements_speculative.txt"),
         # Install Diffusion functionality: pip install angelslim[diffusion]
         "diffusion": get_requirements("requirements/requirements_diffusion.txt"),
-        # Install Diffusion functionality: pip install angelslim[diffusion]
-        "vlm": get_requirements("requirements/requirements_vlm.txt"),
+        # Install multimodal functionality: pip install angelslim[multimodal]
+        "multimodal": get_requirements("requirements/requirements_multimodal.txt"),
     },
     packages=find_packages(),
     python_requires=">=3.0",

From bef484cf22eba2a8d2346f998f409b0abd30ff81 Mon Sep 17 00:00:00 2001
From: ali-12388 <1940747290@qq.com>
Date: Tue, 4 Nov 2025 17:36:24 +0800
Subject: [PATCH 4/6] add fp8_dynamic;add qwen3_omni_quant.md & benchmark

---
 README.md                                     | 35 ++++++++++++-
 README_cn.md                                  | 35 ++++++++++++-
 angelslim/data/multimodal_dataset.py          |  2 +-
 angelslim/utils/config_parser.py              |  2 +-
 ...namic.yaml => qwen3_omni_fp8_dynamic.yaml} |  1 +
 .../fp8_static/qwen3_omni_fp8_static.yaml     |  4 +-
 .../models/qwen3_omni/qwen3_omni_quant.md     | 50 +++++++++++++++++++
 .../performance/quantization/benchmarks.md    | 23 +++++++++
 requirements/requirements.txt                 |  1 -
 tools/run.py                                  |  8 +--
 10 files changed, 150 insertions(+), 11 deletions(-)
 rename configs/qwen3_omni/fp8_dynamic/{qwen3_omn_fp8_dynamic.yaml => qwen3_omni_fp8_dynamic.yaml} (95%)
 create mode 100644 docs/source/models/qwen3_omni/qwen3_omni_quant.md

diff --git a/README.md b/README.md
index 0a6af7d7..daaad81b 100644
--- a/README.md
+++ b/README.md
@@ -510,7 +510,40 @@ Benchmark results for Qwen2.5VL series models with `BF16`、`FP8-Static`、`FP8-
 
 </details>
 
-#### 1.5 Other Models
+#### 1.5 Qwen-Omni Series Models
+
+**Qwen3-Omni Text -> Text Benchmark**
+
+Benchmark results for Qwen3-Omni series models in BF16, FP8-Static, and FP8-Dynamic on aime25, gpqa_diamond, and mmlu_redux are as follows:
+
+<table>
+  <thead>
+    <tr><th>Model</th><th>Quantization</th><th>aime25</th><th>gpqa_diamond</th><th>mmlu_redux</th></tr>
+  </thead>
+  <tbody>
+    <tr><td rowspan="3">Qwen3-Omni-30B-A3B-Instruct</td><td>BF16</td><td>73.32</td><td>56.77</td><td>88.09</td></tr>
+    <tr><td>FP8-Static</td><td>71.33</td><td>56.57</td><td>87.91</td></tr>
+    <tr><td>FP8-Dynamic</td><td>73.33</td><td>55.15</td><td>88.07</td></tr>
+  </tbody>
+</table>
+
+<details>
+<summary>Note</summary>
+
+> - The above evaluation results were obtained by deploying with the vLLM framework and averaging over 5 runs (vLLM only supports the thinker component).
+> - The hyperparameters used during evaluation are as follows:
+> ```json
+>{
+>  "top_p": 0.95,
+>  "temperature": 0.6,
+>  "do_sample": true,
+>  "max-model-len 65536": 65536
+>}
+>```
+
+</details>
+
+#### 1.6 Other Models
 
 Other models such as GLM-4.6, Qwen2.5, and Seed-OSS have been evaluated on benchmarks like `CEVAL`, `MMLU`, and `GSM8K` using quantization strategies including `FP8-Static`, `FP8-Dynamic`, `INT4-GPTQ`, and `INT4-AWQ`.
 
diff --git a/README_cn.md b/README_cn.md
index 0c38b91f..8ca7f723 100644
--- a/README_cn.md
+++ b/README_cn.md
@@ -517,7 +517,40 @@ Qwen2.5VL系列模型的`BF16`、`FP8-Static`、`FP8-Dynamic`、`INT4-GPTQ`、`I
 
 </details>
 
-#### 1.5 其他模型
+#### 1.5 Qwen-Omni 系列模型
+
+**Qwen3-Omni Text -> Text Benchmark**
+
+Qwen3-Omni系列模型的`BF16`、`FP8-Static`、`FP8-Dynamic`在`aime25`、`gpqa_diamond`、`mmlu_redux`上的评测结果如下：
+
+<table>
+  <thead>
+    <tr><th>Model</th><th>Quantization</th><th>aime25</th><th>gpqa_diamond</th><th>mmlu_redux</th></tr>
+  </thead>
+  <tbody>
+    <tr><td rowspan="3">Qwen3-Omni-30B-A3B-Instruct</td><td>BF16</td><td>73.32</td><td>56.77</td><td>88.09</td></tr>
+    <tr><td>FP8-Static</td><td>71.33</td><td>56.57</td><td>87.91</td></tr>
+    <tr><td>FP8-Dynamic</td><td>73.33</td><td>55.15</td><td>88.07</td></tr>
+  </tbody>
+</table>
+
+<details>
+<summary>备注</summary>
+
+> - 以上评测结果使用vllm框架部署测试5次求平均(vllm只支持thinker部分)
+> - 评测时使用的超参如下:
+> ```json
+>{
+>  "top_p": 0.95,
+>  "temperature": 0.6,
+>  "do_sample": true,
+>  "max-model-len 65536": 65536
+>}
+>```
+
+</details>
+
+#### 1.6 其他模型
 
 其他模型比如GLM、Qwen2.5、Seed-OSS等模型利用`FP8-Static`、`FP8-Dynamic`、`INT4-GPTQ`、`INT4-AWQ`量化等策略在`CEVAL`、`MMLU`、`GSM8K`上进行了评测。
 
diff --git a/angelslim/data/multimodal_dataset.py b/angelslim/data/multimodal_dataset.py
index bd0ddc2f..62f42cca 100644
--- a/angelslim/data/multimodal_dataset.py
+++ b/angelslim/data/multimodal_dataset.py
@@ -16,12 +16,12 @@
 import os
 from typing import Dict, List, Union
 
-import qwen_vl_utils
 from datasets import load_dataset
 from PIL import Image
 from tqdm import tqdm
 from transformers import ProcessorMixin
 
+from ..utils.lazy_imports import qwen_vl_utils
 from .base_dataset import BaseDataset
 
 
diff --git a/angelslim/utils/config_parser.py b/angelslim/utils/config_parser.py
index 59e3edfa..c64ee008 100644
--- a/angelslim/utils/config_parser.py
+++ b/angelslim/utils/config_parser.py
@@ -131,6 +131,7 @@ class ModelConfig:
     low_cpu_mem_usage: bool = field(default=True)
     use_cache: bool = field(default=False)
     cache_dir: Optional[str] = field(default=None)
+    use_audio_in_video: bool = field(default=False)
 
 
 @dataclass
@@ -153,7 +154,6 @@ class DatasetConfig:
     num_samples: int = field(default=256)
     batch_size: int = field(default=1)
     shuffle: bool = field(default=False)
-    use_audio_in_video: bool = field(default=False)
     inference_settings: Optional[Dict[str, Any]] = field(default=None)
 
 
diff --git a/configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml b/configs/qwen3_omni/fp8_dynamic/qwen3_omni_fp8_dynamic.yaml
similarity index 95%
rename from configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml
rename to configs/qwen3_omni/fp8_dynamic/qwen3_omni_fp8_dynamic.yaml
index 22e99a89..58b8d748 100644
--- a/configs/qwen3_omni/fp8_dynamic/qwen3_omn_fp8_dynamic.yaml
+++ b/configs/qwen3_omni/fp8_dynamic/qwen3_omni_fp8_dynamic.yaml
@@ -11,6 +11,7 @@ model:
   use_cache: false
   torch_dtype: auto
   device_map: auto
+  use_audio_in_video: false
 
 # Compression configuration
 compression:
diff --git a/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml b/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
index e6f3106a..3c30997c 100644
--- a/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
+++ b/configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
@@ -11,6 +11,7 @@ model:
   use_cache: false
   torch_dtype: auto
   device_map: auto
+  use_audio_in_video: false
 
 # Compression configuration
 compression:
@@ -28,5 +29,4 @@ dataset:
   data_path: your/data/path
   max_seq_length: 8192
   num_samples: 256
-  batch_size: 1
-  use_audio_in_video: false
+  batch_size: 1
\ No newline at end of file
diff --git a/docs/source/models/qwen3_omni/qwen3_omni_quant.md b/docs/source/models/qwen3_omni/qwen3_omni_quant.md
new file mode 100644
index 00000000..2e112545
--- /dev/null
+++ b/docs/source/models/qwen3_omni/qwen3_omni_quant.md
@@ -0,0 +1,50 @@
+# Qwen3-Omni量化指南
+
+Qwen3-Omni模型可采用**FP8（static、dynamic）** 方式进行模型压缩，以下是详细的量化配置与操作说明。
+
+
+## FP8 量化（W8A8）
+
+Qwen3-Omni的FP8量化采用**per-tensor粒度**，支持thinker与talker的llm模块的动态量化（dynamic）和静态量化（static）两种模式。
+
+### 配置参数说明
+
+FP8量化的配置文件可参考路径：`configs/qwen3_omni/fp8_static` 和 `configs/qwen3_omni/fp8_dynamic`，核心参数如下：
+
+#### model配置
+- `name`：模型名称，固定填写`Qwen_Omni`。
+- `model_path`：可填写hugging face模型卡片名称或者本地路径。
+- `use_audio_in_video`: 用于控制是否使用源视频的音频轨道
+
+#### compression配置
+- `name`：压缩策略类型，固定选择量化模式`PTQ`。
+- `quantization.name`：量化算法类型，根据需求选择`fp8_static`（静态量化）或`fp8_dynamic`（动态量化）。
+- `quantization.bits`：量化比特数，FP8量化固定填写`8`。
+- `quantization.quant_method`：权重量化粒度，FP8量化固定为`per-tensor`。
+
+#### dataset配置
+- `name`：数据集类型，固定选择`OmniDataset`。
+- `data_path`：数据集路径，支持jsonl文件路径。自定义数据集需参考`dataset/omni_fake_data/fake_data.json`格式。
+
+### 启动量化流程
+
+通过以下命令启动FP8量化校准：
+
+```shell
+# 动态FP8量化
+python3 tools/run.py -c configs/qwen3_omni/fp8_dynamic/qwen3_omni_fp8_dynamic.yaml
+```
+
+```shell
+# 静态FP8量化
+python3 tools/run.py -c configs/qwen3_omni/fp8_static/qwen3_omni_fp8_static.yaml
+```
+
+## 模型部署
+
+vLLM框架支持Qwen3-Omni的FP8（per-tensor）量化模型部署，建议使用官方部署方式：
+### vllm
+参考[https://github.com/QwenLM/Qwen3-Omni?tab=readme-ov-file#vllm-usage](URL)
+
+### transformers
+参考[https://github.com/QwenLM/Qwen3-Omni?tab=readme-ov-file#transformers-usage](URL)
\ No newline at end of file
diff --git a/docs/source/performance/quantization/benchmarks.md b/docs/source/performance/quantization/benchmarks.md
index 863ee88b..f008705c 100644
--- a/docs/source/performance/quantization/benchmarks.md
+++ b/docs/source/performance/quantization/benchmarks.md
@@ -458,3 +458,26 @@ Qwen3VL系列模型的`BF16`、`FP8-Static`、`FP8-Dynamic`在`MMMU_VAL`、`DocV
 ```
 
 FP8-Dynamic采用Block-wise的量化，启动命令：python3 tools/fp8_quant_blockwise.py --block_size --input_path --output_path
+
+
+## Qwen3-Omni
+
+**Qwen3-Omni Text -> Text Benchmark**
+
+Qwen3-Omni模型的`BF16`、`FP8-Static`、`FP8-Dynamic`在`aime25`、`gpqa_diamond`、`mmlu_redux`上的评测结果如下：
+
+```{eval-rst}
+.. table::
+   :align: center
+   :name: table-qwen3-omni-performance
+
+   +-----------------------------+----------------+----------+--------------+------------+
+   | Model                       | Quantization   | aime25   | gpqa_diamond | mmlu_redux |
+   +=============================+================+==========+==============+============+
+   | Qwen3-Omni-30B-A3B-Instruct | BF16           | 73.32    | 56.77        | 88.09      |
+   +                             +----------------+----------+--------------+------------+
+   |                             | FP8-Static     | 71.33    | 56.57        | 87.91      | 
+   +                             +----------------+----------+--------------+------------+
+   |                             | FP8-Dynamic    | 73.33    | 55.15        | 88.07      | 
+   +-----------------------------+----------------+----------+--------------+------------+
+```
\ No newline at end of file
diff --git a/requirements/requirements.txt b/requirements/requirements.txt
index 47d56103..d093b8c3 100644
--- a/requirements/requirements.txt
+++ b/requirements/requirements.txt
@@ -10,4 +10,3 @@ tiktoken
 datasets
 threadpoolctl
 shortuuid
-qwen_vl_utils==0.0.11
diff --git a/tools/run.py b/tools/run.py
index 0ff5dfc3..8ebb86a5 100644
--- a/tools/run.py
+++ b/tools/run.py
@@ -90,9 +90,9 @@ def multi_nodes_run(config):
         low_cpu_mem_usage=model_config.low_cpu_mem_usage,
         use_cache=model_config.use_cache,
         cache_dir=model_config.cache_dir,
+        use_audio_in_video=model_config.use_audio_in_video,
         deploy_backend=global_config.deploy_backend,
         using_multi_nodes=True,
-        use_audio_in_video=dataset_config.use_audio_in_video,
     )
 
     # Step 5: Prepare data (optional custom dataloader)
@@ -106,7 +106,7 @@ def multi_nodes_run(config):
             num_samples=dataset_config.num_samples,
             shuffle=dataset_config.shuffle,
             inference_settings=dataset_config.inference_settings,
-            use_audio_in_video=dataset_config.use_audio_in_video,
+            use_audio_in_video=model_config.use_audio_in_video,
         )
 
     # Step 6: Initialize compressor
@@ -150,8 +150,8 @@ def run(config):
         low_cpu_mem_usage=model_config.low_cpu_mem_usage,
         use_cache=model_config.use_cache,
         cache_dir=model_config.cache_dir,
+        use_audio_in_video=model_config.use_audio_in_video,
         deploy_backend=global_config.deploy_backend,
-        use_audio_in_video=dataset_config.use_audio_in_video,
     )
 
     # Step 4: Prepare data (optional custom dataloader)
@@ -165,7 +165,7 @@ def run(config):
             num_samples=dataset_config.num_samples,
             shuffle=dataset_config.shuffle,
             inference_settings=dataset_config.inference_settings,
-            use_audio_in_video=dataset_config.use_audio_in_video,
+            use_audio_in_video=model_config.use_audio_in_video,
             model_name=model_config.name,
         )
 

From 81bf38219cf23f5e648a3b8d2e6e0e8b546501ea Mon Sep 17 00:00:00 2001
From: ali-12388 <1940747290@qq.com>
Date: Tue, 4 Nov 2025 18:33:20 +0800
Subject: [PATCH 5/6] modify readme

---
 README.md    | 7 ++++---
 README_cn.md | 7 ++++---
 2 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/README.md b/README.md
index daaad81b..20b032c0 100644
--- a/README.md
+++ b/README.md
@@ -17,7 +17,7 @@ A more accessible, comprehensive, and efficient toolkit for large model compress
 </p>
 
 ## 📣Latest News
-- [25/11/03] We have released v0.2. Quantization support for new models, such as `GLM-4.6` and `Qwen3-VL`, open-sources the Eagle3 speculative decoding training framework, and updates the Diffusion model quantization tools.
+- [25/11/05] We have released v0.2. Quantization support for new models, such as `GLM-4.6` and `Qwen3-VL`, open-sources the Eagle3 speculative decoding training framework, and updates the Diffusion model quantization tools.
 - [25/09/30] We have released **SpecExit**, the reasoning early-exit algorithm: [[Paper]](http://arxiv.org/abs/2509.24248) | [[Docs]](https://angelslim.readthedocs.io/zh-cn/latest/features/speculative_decoding/spec_exit.html) | [[vLLM Code]](https://github.com/vllm-project/vllm/pull/27192)🔥🔥🔥
 - [25/09/26] We have released **TEQUILA**, the ternary quantization algorithm [[Paper]](https://arxiv.org/abs/2509.23809) | [[Code]](https://github.com/Tencent/AngelSlim/tree/tequila/TernaryQuant)🔥🔥🔥
 - [25/09/24] We now support the PTQ quantification of NVFP4 for the Qwen3 series models. We also opensource [Qwen3-32B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-32B_nvfp4) and [Qwen3-235B-A22B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-235B-A22B_nvfp4) weights.
@@ -171,7 +171,8 @@ A more accessible, comprehensive, and efficient toolkit for large model compress
       </td>
       <td>
         <ul style="padding-left: 0; list-style-position: inside;">
-          <li>Under Development</li>
+          <li>FP8-Static</li>
+          <li>FP8-Dynamic</li>
         </ul>
       </td>
       <td>
@@ -512,7 +513,7 @@ Benchmark results for Qwen2.5VL series models with `BF16`、`FP8-Static`、`FP8-
 
 #### 1.5 Qwen-Omni Series Models
 
-**Qwen3-Omni Text -> Text Benchmark**
+**Qwen3-Omni Text to Text Benchmark**
 
 Benchmark results for Qwen3-Omni series models in BF16, FP8-Static, and FP8-Dynamic on aime25, gpqa_diamond, and mmlu_redux are as follows:
 
diff --git a/README_cn.md b/README_cn.md
index 8ca7f723..179eb623 100644
--- a/README_cn.md
+++ b/README_cn.md
@@ -17,7 +17,7 @@
 </p>
 
 ## 📣最新进展
-- [25/11/03] 我们发布V0.2版本，支持了包括GLM-4.6/Qwen3-VL等更多模型的量化，开源投机采样Eagle3训练框架，更新Diffusion模型量化工具。
+- [25/11/05] 我们发布V0.2版本，支持了包括GLM-4.6/Qwen3-VL等更多模型的量化，开源投机采样Eagle3训练框架，更新Diffusion模型量化工具。
 - [25/09/30] 我们开源了思考早退新算法 **SpecExit** [[论文]](http://arxiv.org/abs/2509.24248) | [[文档]](https://angelslim.readthedocs.io/zh-cn/latest/features/speculative_decoding/spec_exit.html) | [[vLLM代码]](https://github.com/vllm-project/vllm/pull/27192)🔥🔥🔥
 - [25/09/30] 我们发布了三值量化新算法 **Tequila** [[论文]](https://arxiv.org/abs/2509.23809) | [[代码]](https://github.com/Tencent/AngelSlim/tree/tequila/TernaryQuant)。🔥🔥🔥
 - [25/09/24] 我们支持了Qwen3系列模型的NVFP4的PTQ量化，我们还开源了[Qwen3-32B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-32B_nvfp4)、[Qwen3-235B-A22B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-235B-A22B_nvfp4)权重。
@@ -172,7 +172,8 @@
       </td>
       <td>
         <ul style="padding-left: 0; list-style-position: inside;">
-          <li>建设中</li>
+          <li>FP8-Static</li>
+          <li>FP8-Dynamic</li>
         </ul>
       </td>
       <td>
@@ -519,7 +520,7 @@ Qwen2.5VL系列模型的`BF16`、`FP8-Static`、`FP8-Dynamic`、`INT4-GPTQ`、`I
 
 #### 1.5 Qwen-Omni 系列模型
 
-**Qwen3-Omni Text -> Text Benchmark**
+**Qwen3-Omni Text to Text Benchmark**
 
 Qwen3-Omni系列模型的`BF16`、`FP8-Static`、`FP8-Dynamic`在`aime25`、`gpqa_diamond`、`mmlu_redux`上的评测结果如下：
 

From e49738f57b2f97be29df0746821c1f7110c6924b Mon Sep 17 00:00:00 2001
From: ali-12388 <1940747290@qq.com>
Date: Tue, 4 Nov 2025 19:10:24 +0800
Subject: [PATCH 6/6] modify readme

---
 README.md                                             | 5 ++---
 README_cn.md                                          | 5 ++---
 angelslim/models/omni/__init__.py                     | 2 +-
 angelslim/models/omni/{qwen_omni.py => qwen3_omni.py} | 0
 docs/source/index.md                                  | 3 ++-
 5 files changed, 7 insertions(+), 8 deletions(-)
 rename angelslim/models/omni/{qwen_omni.py => qwen3_omni.py} (100%)

diff --git a/README.md b/README.md
index 20b032c0..9e4f3f07 100644
--- a/README.md
+++ b/README.md
@@ -17,7 +17,7 @@ A more accessible, comprehensive, and efficient toolkit for large model compress
 </p>
 
 ## 📣Latest News
-- [25/11/05] We have released v0.2. Quantization support for new models, such as `GLM-4.6` and `Qwen3-VL`, open-sources the Eagle3 speculative decoding training framework, and updates the Diffusion model quantization tools.
+- [25/11/05] We have released v0.2. Quantization support for new models, such as `GLM-4.6`, `Qwen3-VL` and `Qwen3-Omni`, open-sources the Eagle3 speculative decoding training framework, and updates the Diffusion model quantization tools.
 - [25/09/30] We have released **SpecExit**, the reasoning early-exit algorithm: [[Paper]](http://arxiv.org/abs/2509.24248) | [[Docs]](https://angelslim.readthedocs.io/zh-cn/latest/features/speculative_decoding/spec_exit.html) | [[vLLM Code]](https://github.com/vllm-project/vllm/pull/27192)🔥🔥🔥
 - [25/09/26] We have released **TEQUILA**, the ternary quantization algorithm [[Paper]](https://arxiv.org/abs/2509.23809) | [[Code]](https://github.com/Tencent/AngelSlim/tree/tequila/TernaryQuant)🔥🔥🔥
 - [25/09/24] We now support the PTQ quantification of NVFP4 for the Qwen3 series models. We also opensource [Qwen3-32B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-32B_nvfp4) and [Qwen3-235B-A22B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-235B-A22B_nvfp4) weights.
@@ -171,8 +171,7 @@ A more accessible, comprehensive, and efficient toolkit for large model compress
       </td>
       <td>
         <ul style="padding-left: 0; list-style-position: inside;">
-          <li>FP8-Static</li>
-          <li>FP8-Dynamic</li>
+          <li><a href="https://github.com/Tencent/AngelSlim/blob/main/docs/source/models/qwen3_omni/qwen3_omni_quant.md">FP8-Static/Dynamic</a></li>
         </ul>
       </td>
       <td>
diff --git a/README_cn.md b/README_cn.md
index 179eb623..cda6f60b 100644
--- a/README_cn.md
+++ b/README_cn.md
@@ -17,7 +17,7 @@
 </p>
 
 ## 📣最新进展
-- [25/11/05] 我们发布V0.2版本，支持了包括GLM-4.6/Qwen3-VL等更多模型的量化，开源投机采样Eagle3训练框架，更新Diffusion模型量化工具。
+- [25/11/05] 我们发布V0.2版本，支持了包括GLM-4.6/Qwen3-VL/Qwen3-Omni等更多模型的量化，开源投机采样Eagle3训练框架，更新Diffusion模型量化工具。
 - [25/09/30] 我们开源了思考早退新算法 **SpecExit** [[论文]](http://arxiv.org/abs/2509.24248) | [[文档]](https://angelslim.readthedocs.io/zh-cn/latest/features/speculative_decoding/spec_exit.html) | [[vLLM代码]](https://github.com/vllm-project/vllm/pull/27192)🔥🔥🔥
 - [25/09/30] 我们发布了三值量化新算法 **Tequila** [[论文]](https://arxiv.org/abs/2509.23809) | [[代码]](https://github.com/Tencent/AngelSlim/tree/tequila/TernaryQuant)。🔥🔥🔥
 - [25/09/24] 我们支持了Qwen3系列模型的NVFP4的PTQ量化，我们还开源了[Qwen3-32B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-32B_nvfp4)、[Qwen3-235B-A22B-NVFP4](https://huggingface.co/AngelSlim/Qwen3-235B-A22B_nvfp4)权重。
@@ -172,8 +172,7 @@
       </td>
       <td>
         <ul style="padding-left: 0; list-style-position: inside;">
-          <li>FP8-Static</li>
-          <li>FP8-Dynamic</li>
+          <li><a href="https://github.com/Tencent/AngelSlim/blob/main/docs/source/models/qwen3_omni/qwen3_omni_quant.md">FP8-Static/Dynamic</a></li>
         </ul>
       </td>
       <td>
diff --git a/angelslim/models/omni/__init__.py b/angelslim/models/omni/__init__.py
index 2dfd628f..6fb9571a 100644
--- a/angelslim/models/omni/__init__.py
+++ b/angelslim/models/omni/__init__.py
@@ -13,4 +13,4 @@
 # limitations under the License.
 
 
-from .qwen_omni import Qwen_Omni  # noqa: F401
+from .qwen3_omni import Qwen_Omni  # noqa: F401
diff --git a/angelslim/models/omni/qwen_omni.py b/angelslim/models/omni/qwen3_omni.py
similarity index 100%
rename from angelslim/models/omni/qwen_omni.py
rename to angelslim/models/omni/qwen3_omni.py
diff --git a/docs/source/index.md b/docs/source/index.md
index 305a8dac..b5a0e08e 100644
--- a/docs/source/index.md
+++ b/docs/source/index.md
@@ -57,7 +57,7 @@ AngelSlim是腾讯自研的，致力于打造更易用、更全面和更高效
    |                   | Wan             |                      |            | 建设中          |
    |                   | SDXL            |                      |            |                 |
    +-------------------+-----------------+----------------------+------------+-----------------+
-   | **语音(TTS/ASR)** | Qwen3-Omni      | 建设中                 | 建设中     | **Token剪枝**   |
+   | **语音(TTS/ASR)** | Qwen3-Omni      | FP8-Static/Dynamic    | 建设中     | **Token剪枝**   |
    |                   |                 |                      |            | 建设中          |
    +-------------------+-----------------+----------------------+------------+-----------------+
 
@@ -94,6 +94,7 @@ models/hunyuan/hunyuan_quant
 models/deepseek/deepseek_quant
 models/qwen/qwen_quant
 models/qwenvl/qwenvl_quant
+models/qwen3_omni/qwen3_omni_quant
 :::