mediapipe-nightly 0.10.11.post20240301__cp311-cp311-macosx_11_0_x86_64.whl → 0.10.11.post20240302__cp311-cp311-macosx_11_0_x86_64.whl

mediapipe/__init__.py

@@ -23,4 +23,4 @@ del modules
 del python
 del mediapipe
 del util
-__version__ = '0.10.11-20240301'
+__version__ = '0.10.11-20240302'

mediapipe/tasks/cc/genai/inference/calculators/detokenizer_calculator_pb2.py

@@ -0,0 +1,27 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# source: mediapipe/tasks/cc/genai/inference/calculators/detokenizer_calculator.proto
+# Protobuf Python Version: 4.25.1
+"""Generated protocol buffer code."""
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import symbol_database as _symbol_database
+from google.protobuf.internal import builder as _builder
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\nKmediapipe/tasks/cc/genai/inference/calculators/detokenizer_calculator.proto\x12\x10odml.infra.proto\"q\n\x1c\x44\x65tokenizerCalculatorOptions\x12\x16\n\x0espm_model_file\x18\x01 \x01(\t\x12\x13\n\x0bstop_tokens\x18\x04 \x03(\t\x12\x18\n\x10num_output_heads\x18\x05 \x01(\x05J\x04\x08\x02\x10\x03J\x04\x08\x03\x10\x04\x42@\n\x1b\x63om.google.odml.infra.protoB!DetokenizerCalculatorOptionsProtob\x06proto3')
+
+_globals = globals()
+_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
+_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'mediapipe.tasks.cc.genai.inference.calculators.detokenizer_calculator_pb2', _globals)
+if _descriptor._USE_C_DESCRIPTORS == False:
+  _globals['DESCRIPTOR']._options = None
+  _globals['DESCRIPTOR']._serialized_options = b'\n\033com.google.odml.infra.protoB!DetokenizerCalculatorOptionsProto'
+  _globals['_DETOKENIZERCALCULATOROPTIONS']._serialized_start=97
+  _globals['_DETOKENIZERCALCULATOROPTIONS']._serialized_end=210
+# @@protoc_insertion_point(module_scope)

mediapipe/tasks/cc/genai/inference/calculators/llm_gpu_calculator_pb2.py

@@ -0,0 +1,31 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# source: mediapipe/tasks/cc/genai/inference/calculators/llm_gpu_calculator.proto
+# Protobuf Python Version: 4.25.1
+"""Generated protocol buffer code."""
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import symbol_database as _symbol_database
+from google.protobuf.internal import builder as _builder
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+from mediapipe.tasks.cc.genai.inference.proto import llm_file_metadata_pb2 as mediapipe_dot_tasks_dot_cc_dot_genai_dot_inference_dot_proto_dot_llm__file__metadata__pb2
+from mediapipe.tasks.cc.genai.inference.proto import llm_params_pb2 as mediapipe_dot_tasks_dot_cc_dot_genai_dot_inference_dot_proto_dot_llm__params__pb2
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\nGmediapipe/tasks/cc/genai/inference/calculators/llm_gpu_calculator.proto\x12\x10odml.infra.proto\x1a@mediapipe/tasks/cc/genai/inference/proto/llm_file_metadata.proto\x1a\x39mediapipe/tasks/cc/genai/inference/proto/llm_params.proto\"\xf4\x04\n\x17LlmGpuCalculatorOptions\x12\x13\n\x0bweight_path\x18\x01 \x01(\t\x12N\n\x0egpu_model_info\x18\n \x01(\x0b\x32\x36.odml.infra.proto.LlmGpuCalculatorOptions.GpuModelInfo\x12\x19\n\x11num_decode_tokens\x18\x0c \x01(\x05\x12\x1b\n\x13sequence_batch_size\x18\x0e \x01(\x05\x12\x11\n\tlora_path\x18\x13 \x01(\t\x12\x37\n\x0ellm_parameters\x18\x14 \x01(\x0b\x32\x1f.odml.infra.proto.LlmParameters\x12\x18\n\x10num_output_heads\x18\x16 \x01(\x05\x12\x0c\n\x04topk\x18\x17 \x01(\x05\x12\x13\n\x0btemperature\x18\x18 \x01(\x02\x12\x18\n\x0brandom_seed\x18\x19 \x01(\rH\x00\x88\x01\x01\x1a\xc0\x01\n\x0cGpuModelInfo\x12\x1c\n\x14\x61llow_precision_loss\x18\x01 \x01(\x08\x12\x1a\n\x12\x65nable_fast_tuning\x18\x02 \x01(\x08\x12\x1b\n\x13\x65nable_winograd_opt\x18\x03 \x01(\x08\x12\x15\n\ruse_low_power\x18\x04 \x01(\x08\x12\x1e\n\x16prefer_texture_weights\x18\x05 \x01(\x08\x12\"\n\x1a\x65nable_host_mapped_pointer\x18\x06 \x01(\x08\x42\x0e\n\x0c_random_seedJ\x04\x08\x02\x10\x03J\x04\x08\x03\x10\x04J\x04\x08\x04\x10\x05J\x04\x08\x05\x10\x06J\x04\x08\x06\x10\x07J\x04\x08\x07\x10\x08J\x04\x08\x08\x10\tJ\x04\x08\t\x10\nJ\x04\x08\x0b\x10\x0cJ\x04\x08\r\x10\x0eJ\x04\x08\x1a\x10\x1bJ\x04\x08\x1b\x10\x1c\x42;\n\x1b\x63om.google.odml.infra.protoB\x1cLlmGpuCalculatorOptionsProtob\x06proto3')
+
+_globals = globals()
+_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
+_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'mediapipe.tasks.cc.genai.inference.calculators.llm_gpu_calculator_pb2', _globals)
+if _descriptor._USE_C_DESCRIPTORS == False:
+  _globals['DESCRIPTOR']._options = None
+  _globals['DESCRIPTOR']._serialized_options = b'\n\033com.google.odml.infra.protoB\034LlmGpuCalculatorOptionsProto'
+  _globals['_LLMGPUCALCULATOROPTIONS']._serialized_start=219
+  _globals['_LLMGPUCALCULATOROPTIONS']._serialized_end=847
+  _globals['_LLMGPUCALCULATOROPTIONS_GPUMODELINFO']._serialized_start=567
+  _globals['_LLMGPUCALCULATOROPTIONS_GPUMODELINFO']._serialized_end=759
+# @@protoc_insertion_point(module_scope)

mediapipe/tasks/cc/genai/inference/calculators/tokenizer_calculator_pb2.py

@@ -0,0 +1,29 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# source: mediapipe/tasks/cc/genai/inference/calculators/tokenizer_calculator.proto
+# Protobuf Python Version: 4.25.1
+"""Generated protocol buffer code."""
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import symbol_database as _symbol_database
+from google.protobuf.internal import builder as _builder
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\nImediapipe/tasks/cc/genai/inference/calculators/tokenizer_calculator.proto\x12\x10odml.infra.proto\"\xb7\x02\n\x1aTokenizerCalculatorOptions\x12\x12\n\nmax_tokens\x18\x01 \x01(\x05\x12\x18\n\x0espm_model_file\x18\x02 \x01(\tH\x00\x12Y\n\x11tflite_model_file\x18\x04 \x01(\x0b\x32<.odml.infra.proto.TokenizerCalculatorOptions.TfLiteModelFileH\x00\x12\x16\n\x0estart_token_id\x18\x03 \x01(\x05\x12 \n\x18\x62ytes_to_unicode_mapping\x18\x05 \x01(\x08\x1aH\n\x0fTfLiteModelFile\x12\x12\n\nmodel_file\x18\x01 \x01(\t\x12!\n\x19spm_model_key_in_metadata\x18\x02 \x01(\tB\x0c\n\nmodel_fileB>\n\x1b\x63om.google.odml.infra.protoB\x1fTokenizerCalculatorOptionsProtob\x06proto3')
+
+_globals = globals()
+_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
+_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'mediapipe.tasks.cc.genai.inference.calculators.tokenizer_calculator_pb2', _globals)
+if _descriptor._USE_C_DESCRIPTORS == False:
+  _globals['DESCRIPTOR']._options = None
+  _globals['DESCRIPTOR']._serialized_options = b'\n\033com.google.odml.infra.protoB\037TokenizerCalculatorOptionsProto'
+  _globals['_TOKENIZERCALCULATOROPTIONS']._serialized_start=96
+  _globals['_TOKENIZERCALCULATOROPTIONS']._serialized_end=407
+  _globals['_TOKENIZERCALCULATOROPTIONS_TFLITEMODELFILE']._serialized_start=321
+  _globals['_TOKENIZERCALCULATOROPTIONS_TFLITEMODELFILE']._serialized_end=393
+# @@protoc_insertion_point(module_scope)

mediapipe/tasks/cc/genai/inference/proto/llm_file_metadata_pb2.py

@@ -0,0 +1,32 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# source: mediapipe/tasks/cc/genai/inference/proto/llm_file_metadata.proto
+# Protobuf Python Version: 4.25.1
+"""Generated protocol buffer code."""
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import symbol_database as _symbol_database
+from google.protobuf.internal import builder as _builder
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+from mediapipe.tasks.cc.genai.inference.proto import llm_params_pb2 as mediapipe_dot_tasks_dot_cc_dot_genai_dot_inference_dot_proto_dot_llm__params__pb2
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n@mediapipe/tasks/cc/genai/inference/proto/llm_file_metadata.proto\x12\x10odml.infra.proto\x1a\x39mediapipe/tasks/cc/genai/inference/proto/llm_params.proto\"\xe8\x02\n\x0fLlmFileMetadata\x12=\n\x07tensors\x18\x01 \x03(\x0b\x32,.odml.infra.proto.LlmFileMetadata.TensorInfo\x12\x35\n\x0cmodel_params\x18\x02 \x01(\x0b\x32\x1f.odml.infra.proto.LlmParameters\x12\x11\n\tlora_rank\x18\x03 \x01(\x05\x1a\xcb\x01\n\nTensorInfo\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x0e\n\x06offset\x18\x02 \x01(\x04\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12H\n\tdata_type\x18\x04 \x01(\x0e\x32\x35.odml.infra.proto.LlmFileMetadata.TensorInfo.DataType\"G\n\x08\x44\x61taType\x12\x0f\n\x0bUNSPECIFIED\x10\x00\x12\x0b\n\x07\x46LOAT32\x10\x01\x12\x08\n\x04INT8\x10\x02\x12\x08\n\x04INT4\x10\x03\x12\t\n\x05UINT4\x10\x04\x42\x33\n\x1b\x63om.google.odml.infra.protoB\x14LlmFileMetadataProtob\x06proto3')
+
+_globals = globals()
+_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
+_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'mediapipe.tasks.cc.genai.inference.proto.llm_file_metadata_pb2', _globals)
+if _descriptor._USE_C_DESCRIPTORS == False:
+  _globals['DESCRIPTOR']._options = None
+  _globals['DESCRIPTOR']._serialized_options = b'\n\033com.google.odml.infra.protoB\024LlmFileMetadataProto'
+  _globals['_LLMFILEMETADATA']._serialized_start=146
+  _globals['_LLMFILEMETADATA']._serialized_end=506
+  _globals['_LLMFILEMETADATA_TENSORINFO']._serialized_start=303
+  _globals['_LLMFILEMETADATA_TENSORINFO']._serialized_end=506
+  _globals['_LLMFILEMETADATA_TENSORINFO_DATATYPE']._serialized_start=435
+  _globals['_LLMFILEMETADATA_TENSORINFO_DATATYPE']._serialized_end=506
+# @@protoc_insertion_point(module_scope)

mediapipe/tasks/python/genai/converter/converter_base.py

@@ -14,6 +14,7 @@
 
 """Defines a couple base classes for the conversion/quantization process."""
 
+from typing import Iterator
 import os
 from typing import Dict, List, Optional, Tuple
 import numpy as np
@@ -105,7 +106,9 @@ class CkptLoaderBase:
     self._feedforward_quant_bits = feedforward_quant_bits
     self._embedding_quant_bits = embedding_quant_bits
 
-  def load_to_actions(self) -> List[Optional[QuantizationAction]]:
+  def load_to_actions(
+      self,
+  ) -> Iterator[Optional[List[QuantizationAction]]]:
     """Loads the checkpoint and returns the quantization actions."""
     raise NotImplementedError("The load_to_actions method is not implemented.")
 

mediapipe/tasks/python/genai/converter/llm_converter.py

@@ -193,18 +193,21 @@ def convert_checkpoint(config: ConversionConfig) -> None:
     )
     actions = loader.load_to_actions()
 
-    # Quantize the weights.
-    quantized_tensors = quantize_by_actions(
-        actions, config.backend, config.is_symmetric
-    )
-
-    # Write the quantized tensors into file(s).
-    writer = converter_factory.create_writer(
-        writer_type='weight_bins',
-        output_dir=config.output_dir,
-        backend=config.backend,
-    )
-    writer.write_variables(quantized_tensors)
+    for action in actions:
+      # Quantize the weight.
+      quantized_tensors = quantize_by_actions(
+          action, config.backend, config.is_symmetric
+      )
+      del action
+      # Write the quantized tensors into file(s).
+      writer = converter_factory.create_writer(
+          writer_type='weight_bins',
+          output_dir=config.output_dir,
+          backend=config.backend,
+      )
+      writer.write_variables(quantized_tensors)
+      del quantized_tensors
+      del writer
 
   combined_weight_bins_to_tflite(
       config.model_type,

mediapipe/tasks/python/genai/converter/pytorch_converter.py

@@ -14,6 +14,7 @@
 
 """CkptLoader implementation for loading the Pytorch file."""
 
+from typing import Iterator
 import enum
 import os
 from typing import List, Optional
@@ -306,10 +307,12 @@ class PytorchCkptLoader(converter_base.CkptLoaderBase):
     else:
       raise ValueError(f"Unknown special model: {special_model}")
 
-  def load_to_actions(self):
+  def load_to_actions(
+      self,
+  ) -> Iterator[List[converter_base.QuantizationAction]]:
     tensor_names = self._reader.get_tensor_names()
-    actions = []
     for tensor_name in tensor_names:
       tensor_actions = self.mapper.map_to_actions(tensor_name)
-      actions.extend(tensor_actions)
-    return actions
+      if tensor_actions is None:
+        continue
+      yield tensor_actions

mediapipe/tasks/python/genai/converter/pytorch_converter_test.py

@@ -79,7 +79,7 @@ class PytorchConverterTest(parameterized.TestCase):
     actions = loader.load_to_actions()
     # There are 16 layers in the model, but qkv weight and bias would be
     # decomposed to q, k, v tensors, so there would be 20 quantization actions.
-    self.assertLen(actions, 20)
+    self.assertEqual(sum(len(action) for action in actions), 20)
 
 
 if __name__ == '__main__':

mediapipe/tasks/python/genai/converter/safetensors_converter.py

@@ -15,6 +15,7 @@
 """CkptLoader implementation for loading the Safetensors."""
 
 import array
+from typing import Iterator
 import enum
 import glob
 import json
@@ -510,12 +511,12 @@ class SafetensorsCkptLoader(converter_base.CkptLoaderBase):
     else:
       raise ValueError(f"Unknown special model: {special_model}")
 
-  def load_to_actions(self) -> List[converter_base.QuantizationAction]:
+  def load_to_actions(
+      self,
+  ) -> Iterator[List[converter_base.QuantizationAction]]:
     tensor_names = self._reader.get_tensor_names()
-    actions = []
     for tensor_name in tensor_names:
       tensor_actions = self.mapper.map_to_actions(tensor_name)
       if tensor_actions is None:
         continue
-      actions.extend(tensor_actions)
-    return actions
+      yield tensor_actions

mediapipe/tasks/python/genai/converter/safetensors_converter_test.py

@@ -76,7 +76,7 @@ class SafetensorsConverterTest(parameterized.TestCase):
         backend='gpu',
     )
     actions = loader.load_to_actions()
-    self.assertLen(actions, 15)
+    self.assertLen(list(actions), 15)
 
 
 if __name__ == '__main__':

{mediapipe_nightly-0.10.11.post20240301.dist-info → mediapipe_nightly-0.10.11.post20240302.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mediapipe-nightly
-Version: 0.10.11.post20240301
+Version: 0.10.11.post20240302
 Summary: MediaPipe is the simplest way for researchers and developers to build world-class ML solutions and applications for mobile, edge, cloud and the web.
 Home-page: https://github.com/google/mediapipe
 Author: The MediaPipe Authors

{mediapipe_nightly-0.10.11.post20240301.dist-info → mediapipe_nightly-0.10.11.post20240302.dist-info}/RECORD

@@ -1,9 +1,9 @@
-mediapipe_nightly-0.10.11.post20240301.dist-info/RECORD,,
-mediapipe_nightly-0.10.11.post20240301.dist-info/LICENSE,sha256=hwfu8FM5h-_FsVXWR2HutuIHk_ULm9Gmja0c9HGdDtg,12331
-mediapipe_nightly-0.10.11.post20240301.dist-info/WHEEL,sha256=_dGy9EKu13jbT4E4ph0EWQuZFac2dCXB7_Dw2ZQDLNI,111
-mediapipe_nightly-0.10.11.post20240301.dist-info/top_level.txt,sha256=LG-epD1oIiiHFRqLp--7jacjB3dbx2RfMcLYjCIhmxU,175
-mediapipe_nightly-0.10.11.post20240301.dist-info/METADATA,sha256=nai98QxgVWO7p-hpRQM56n9JiHhMR8VPbVvv7_vS83c,9715
-mediapipe/__init__.py,sha256=1ug3U5bLdO0GSoC7AftR3EItV8On2NyiO1Hpg0_oqy0,816
+mediapipe_nightly-0.10.11.post20240302.dist-info/RECORD,,
+mediapipe_nightly-0.10.11.post20240302.dist-info/LICENSE,sha256=hwfu8FM5h-_FsVXWR2HutuIHk_ULm9Gmja0c9HGdDtg,12331
+mediapipe_nightly-0.10.11.post20240302.dist-info/WHEEL,sha256=_dGy9EKu13jbT4E4ph0EWQuZFac2dCXB7_Dw2ZQDLNI,111
+mediapipe_nightly-0.10.11.post20240302.dist-info/top_level.txt,sha256=LG-epD1oIiiHFRqLp--7jacjB3dbx2RfMcLYjCIhmxU,175
+mediapipe_nightly-0.10.11.post20240302.dist-info/METADATA,sha256=Z_ixYoKtEPje2xOmQuHu44KTR17vwvA_rhw1LA27h1E,9715
+mediapipe/__init__.py,sha256=FG3p8a-sPsr3_nru1WF1PkqKrAHldPzVed4uOfnptR8,816
 mediapipe/tasks/__init__.py,sha256=sVJS2p8J2PNVl8DLRPVY6KLpHenP_z3QVPRU0x_iL5g,571
 mediapipe/tasks/python/__init__.py,sha256=wIM_WOWboOVI1MeehN8fkN_DjoA0MEBVw5mShAd8AS4,858
 mediapipe/tasks/python/benchmark/__init__.py,sha256=epEucluzX0HinwBZoS7Tgb19j_qgfTuBf-vBkqemch8,587
@@ -86,16 +86,16 @@ mediapipe/tasks/python/text/text_classifier.py,sha256=AJbYep6iL8vkf6JKRrGArr9sNd
 mediapipe/tasks/python/text/core/base_text_task_api.py,sha256=OHt7j_0n5c3HBdOrCb_BGWCdKWMKvDULp6tKA5mDZAc,1822
 mediapipe/tasks/python/text/core/__init__.py,sha256=ZKC2XRtShVe6k6u6LxDt1pG7DQIn5nZnjurs6Pcvm6A,593
 mediapipe/tasks/python/genai/__init__.py,sha256=7rri6fT6wNurla8O2c5yKiLs9_3qIY0vKkyVAUDe-18,620
-mediapipe/tasks/python/genai/converter/pytorch_converter_test.py,sha256=hJOjozkWpzjedhkfWoF-vinSMs7TqCPO-W9VpleQo2U,3008
-mediapipe/tasks/python/genai/converter/safetensors_converter_test.py,sha256=bdBHCQOTqAyHpC19di_H0hV3zraBOXXfV_6NbFT0ouw,2796
+mediapipe/tasks/python/genai/converter/pytorch_converter_test.py,sha256=y_Mg9pOQtlUDh6uVmkz5LcUbk-pmDLA9L3KcxKR-OaA,3041
+mediapipe/tasks/python/genai/converter/safetensors_converter_test.py,sha256=oCk4FnsjBJkEPlXtv8fdq9dn3I06LsSQRMi0BV_9mew,2802
 mediapipe/tasks/python/genai/converter/quantization_util.py,sha256=B6i13GqRRIwMabEJWO8rFHPMBjIgdOhFpHiwMD4GzRc,17196
-mediapipe/tasks/python/genai/converter/safetensors_converter.py,sha256=IbnvTH_n8hnVd9nChOT8oGx2Ssw4iCFCnlWB1708ujc,18710
+mediapipe/tasks/python/genai/converter/safetensors_converter.py,sha256=ttCpqyMeHkIFzppP8G8uMG5VsslK5-ULFT5U3R_qif0,18713
 mediapipe/tasks/python/genai/converter/weight_bins_writer_test.py,sha256=6qgNYXODNOsbveZ0ighEW4JBdawil9mPcC16MZ0mdm8,1994
-mediapipe/tasks/python/genai/converter/converter_base.py,sha256=nRIhPYUg2pkUZcmwMIDVM4lskRRUwoOYep9MY78b2xE,6519
-mediapipe/tasks/python/genai/converter/pytorch_converter.py,sha256=RiuCr2aXbZVn3Ssa3OcSx_RQK56rfZRCpvG7qYIji4w,10675
+mediapipe/tasks/python/genai/converter/converter_base.py,sha256=1nBgvcY5xaI0ZPjBJTVdViDFraVyJUZHLXBT0dYOX9c,6568
+mediapipe/tasks/python/genai/converter/pytorch_converter.py,sha256=b-GWYOzgD-ZRGgyqcXE9LG_JOL0Mqba4q0pc_imrbrg,10771
 mediapipe/tasks/python/genai/converter/__init__.py,sha256=jfUkinDJR5BVldnbJMbo5vIr2Xc5Z4TTnaCJTNoAUvg,893
 mediapipe/tasks/python/genai/converter/converter_factory.py,sha256=2K16PZBQym0WhXM2HOdBMHMugykohoD4OTaOIo-UKko,2928
-mediapipe/tasks/python/genai/converter/llm_converter.py,sha256=AFX75Eay438uKNEt0N5rYyNB9FXQ3Sr-7GnuJvGryvM,7788
+mediapipe/tasks/python/genai/converter/llm_converter.py,sha256=OJqrYEDPQYwoOftBw-Ajsqtkcj9uRHBAb_Oo_RCEKa4,7896
 mediapipe/tasks/python/genai/converter/quantization_util_test.py,sha256=ICujhTFeREGuHGmNk1PlBpf1AUThFvv-Wl5UuZ-xWAk,9060
 mediapipe/tasks/python/genai/converter/weight_bins_writer.py,sha256=qrsjTWw99u-VDNhtHbnbDfMypx4sGfGyY8mMBXYnQtA,4347
 mediapipe/tasks/python/metadata/metadata.py,sha256=EECQnM-Af0angD60jaBBOuNMgt7HExH6SqVtVMFNHGc,33763
@@ -258,10 +258,15 @@ mediapipe/tasks/cc/genai/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZ
 mediapipe/tasks/cc/genai/inference/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mediapipe/tasks/cc/genai/inference/proto/llm_params_pb2.py,sha256=zGZQREcWs5jzbMipVcGmXOrJFMeq5lgQNcA_U1XkOoE,2311
 mediapipe/tasks/cc/genai/inference/proto/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+mediapipe/tasks/cc/genai/inference/proto/llm_file_metadata_pb2.py,sha256=MdDe96Ruxqcnm4vrixmyLqPmBWpXn207W50TX5UYMCo,2454
 mediapipe/tasks/cc/genai/inference/proto/transformer_params_pb2.py,sha256=-KHfCIuevXnBJusJ2o3_jaf4GHSq0no-o1BhR2jRPTk,5080
 mediapipe/tasks/cc/genai/inference/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mediapipe/tasks/cc/genai/inference/utils/llm_utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mediapipe/tasks/cc/genai/inference/utils/xnn_utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+mediapipe/tasks/cc/genai/inference/calculators/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+mediapipe/tasks/cc/genai/inference/calculators/tokenizer_calculator_pb2.py,sha256=m6mFdMkEGTeGJOa5o00kQU7rVZRDmnBhum0kfEmQJ_8,2038
+mediapipe/tasks/cc/genai/inference/calculators/llm_gpu_calculator_pb2.py,sha256=vm7Oi2HiCZLI1fx6GDUOgfLcwfKebMGZOk3nFl_X5R0,3191
+mediapipe/tasks/cc/genai/inference/calculators/detokenizer_calculator_pb2.py,sha256=-VpQshysnt83CsxDB5AIUEb9qw6KaArkek1MR7xbJxI,1625
 mediapipe/tasks/cc/genai/inference/c/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mediapipe/tasks/cc/metadata/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mediapipe/tasks/cc/metadata/python/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -314,7 +319,7 @@ mediapipe/util/sequence/media_sequence_util_test.py,sha256=nuN9-HW3kw2kZbraCH76q
 mediapipe/util/analytics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mediapipe/util/analytics/mediapipe_log_extension_pb2.py,sha256=UX1x8jckc8NQbUz-SXOKkb8co3JNIRX3rfaMs4_Dz8g,4319
 mediapipe/util/analytics/mediapipe_logging_enums_pb2.py,sha256=9pxs-DNSQnXQe7E_LKokGDbD_G1FpmzzJw1jzex78lU,3781
-mediapipe/python/_framework_bindings.cpython-311-darwin.so,sha256=8ZEYR0i-fkiOBswC7tu40j59fxqfb5swrlYDElBo1HY,78943555
+mediapipe/python/_framework_bindings.cpython-311-darwin.so,sha256=IlYGT68y4YVPsQx8ZeHDSDPeDoyzQcl2D0hen0QWmxc,78906003
 mediapipe/python/solution_base.py,sha256=nEIqsho9DlutfvWWzdSxCOpJ2QzN7n2938WLDmFzn38,26072
 mediapipe/python/timestamp_test.py,sha256=oWKTZMsV586jH57OBV30rihcymETyGC29VbYURNLJQQ,2528
 mediapipe/python/image_frame_test.py,sha256=ZSjdE-an2t8i6MiA4_Xri91VMH5_CCx45fjhWUQptMY,8602