litert-torch-nightly 0.9.0.dev20260130__py3-none-any.whl → 0.9.0.dev20260131__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -14,6 +14,7 @@
14
14
  # ==============================================================================
15
15
  """Export library for HF integration."""
16
16
 
17
+ import gc
17
18
  import os
18
19
  import time
19
20
 
@@ -258,8 +259,18 @@ def export_text_prefill_decode_model(
258
259
  elapsed_time = end_time - start_time
259
260
  print(f'Model conversion executed in {elapsed_time} seconds.')
260
261
 
262
+ del lrt_model
263
+ del converter
264
+ gc.collect()
265
+
261
266
  # Quantization
262
- return maybe_quantize_model(model_path, quantization_recipe)
267
+ quantization_recipe_list = (
268
+ quantization_recipe.split(',') if quantization_recipe else [None]
269
+ )
270
+ for recipe in quantization_recipe_list:
271
+ model_path = maybe_quantize_model(model_path, recipe)
272
+ gc.collect()
273
+ return model_path
263
274
 
264
275
 
265
276
  def maybe_quantize_model(
@@ -271,7 +282,8 @@ def maybe_quantize_model(
271
282
  return model_path
272
283
  start_time = time.perf_counter()
273
284
  quantized_model_path = (
274
- model_path.removesuffix('.tflite') + '_quantized.tflite'
285
+ model_path.removesuffix('.tflite').removesuffix('_quantized')
286
+ + '_quantized.tflite'
275
287
  )
276
288
  qt = quantizer_lib.Quantizer(model_path)
277
289
  try:
@@ -316,7 +328,13 @@ def export_embedder_model(
316
328
  lrt_model = converter.convert(strict_export=False)
317
329
  model_path = os.path.join(work_dir, 'model.tflite')
318
330
  lrt_model.export(model_path)
319
- return maybe_quantize_model(model_path, quantization_recipe)
331
+ quantization_recipe_list = (
332
+ quantization_recipe.split(',') if quantization_recipe else [None]
333
+ )
334
+ for recipe in quantization_recipe_list:
335
+ model_path = maybe_quantize_model(model_path, recipe)
336
+ gc.collect()
337
+ return model_path
320
338
 
321
339
 
322
340
  def export_tokenizer(
@@ -14,6 +14,7 @@
14
14
  # ==============================================================================
15
15
  """Export functions for HuggingFace Transformers models."""
16
16
 
17
+ import gc
17
18
  import os
18
19
 
19
20
  from litert_torch.generative.export_hf.core import export_lib
@@ -60,6 +61,7 @@ def export(
60
61
  export_lib.export_text_prefill_decode_model(
61
62
  pt_model, text_model_config, export_config, work_dir, quantization_recipe
62
63
  )
64
+ gc.collect()
63
65
  tokenizer_model_path = export_lib.export_tokenizer(tokenizer, work_dir)
64
66
  tflite_model_path = os.path.join(
65
67
  work_dir,
litert_torch/version.py CHANGED
@@ -15,4 +15,4 @@
15
15
 
16
16
  # The next version of litert-torch.
17
17
  # The minor version code should be bumped after every release.
18
- __version__ = "0.9.0.dev20260130"
18
+ __version__ = "0.9.0.dev20260131"
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: litert-torch-nightly
3
- Version: 0.9.0.dev20260130
3
+ Version: 0.9.0.dev20260131
4
4
  Summary: Support PyTorch model conversion with LiteRT.
5
5
  Home-page: https://github.com/google-ai-edge/litert-torch
6
6
  Keywords: On-Device ML,AI,Google,TFLite,LiteRT,PyTorch,LLMs,GenAI
@@ -3,7 +3,7 @@ litert_torch/_config.py,sha256=zDnki83sBsQzDAea6bvzwccylWHnPUzbEyGGRh6B14w,2526
3
3
  litert_torch/cli.py,sha256=TiAUbgbWm3ecTUJtJ1_hjKJuC1LrG-Qwnm8_zws-sVY,984
4
4
  litert_torch/conftest.py,sha256=gYmFrsR4c_fjIidbyrDnek26yS0crDP6-UoyMvy-WFg,757
5
5
  litert_torch/model.py,sha256=KXFTyyfPM6AnP0JoSwsTqQR3lUQbMkTGSr3dUsfQ5Jk,5635
6
- litert_torch/version.py,sha256=6AkZDFuaj34uiTiNVFv7ms5Cr4xOn8DAWvzn3tWt5J4,804
6
+ litert_torch/version.py,sha256=Pn4ldfbr-sKdhdoRgZKENgSSbOvnfObW9NuvYwVmd7o,804
7
7
  litert_torch/_convert/__init__.py,sha256=qdLdbj5NjhNG-QgY5O_8TzOr2XaDoWvmdY9JNPStQmw,670
8
8
  litert_torch/_convert/conversion.py,sha256=NuQEphyYp3W19IKvyTWo9pe7zt1-XmWM4zU9PDkUm54,6108
9
9
  litert_torch/_convert/conversion_utils.py,sha256=MWpB-3eN-rvQzTtXsPL30cDIK431SQuwvw3ia2K2ONM,2158
@@ -179,7 +179,7 @@ litert_torch/generative/examples/tiny_llama/verify.py,sha256=6geA8OUOSj8_sTRyoo0
179
179
  litert_torch/generative/examples/tiny_llama/verify_util.py,sha256=FKMC6Olex6bJbB8HXvC1KwxPbKgRBfT1CjoWcmyaPD8,2989
180
180
  litert_torch/generative/export_hf/__init__.py,sha256=5xWIp2ziIwapcZcjSKfeaFgBnIooa8ckhTQ7mazZC3c,670
181
181
  litert_torch/generative/export_hf/__main__.py,sha256=8VuBDkZ2sL-q2XdQ45qwzeHQk39-MM_6TdkxOU_23xE,782
182
- litert_torch/generative/export_hf/export.py,sha256=FGjfwyWBNzGaBxTZHq-w0qq1W9BG0SuM1OV67G8nmfI,2686
182
+ litert_torch/generative/export_hf/export.py,sha256=BIgdZxD3BhJ5ImZkwGAi7mSqyspbafx_AcYwMMooD_U,2711
183
183
  litert_torch/generative/export_hf/export_main.py,sha256=bQidNXz0MEP_gil86LSfnpCW0pUiqZq2-F9ZOrSb3Yk,1183
184
184
  litert_torch/generative/export_hf/core/__init__.py,sha256=5xWIp2ziIwapcZcjSKfeaFgBnIooa8ckhTQ7mazZC3c,670
185
185
  litert_torch/generative/export_hf/core/attention.py,sha256=FFqq5TZ-V4Bdn2jb06_VWjAzVj1dlDuV1Llkftf_J54,2454
@@ -187,7 +187,7 @@ litert_torch/generative/export_hf/core/attention_test.py,sha256=RevOczfPncmbIBth
187
187
  litert_torch/generative/export_hf/core/cache.py,sha256=pfWh2SACdhNY2of2Z8KJC0wrSQ2jrkXgPHWe7PSEiuU,10263
188
188
  litert_torch/generative/export_hf/core/cache_base.py,sha256=FXMm9B8nDwC8uTyLmuBnYKLTnNtoeGN8gUnWwDCcH08,1714
189
189
  litert_torch/generative/export_hf/core/cache_test.py,sha256=y-v-oOGtRNPGWRfIfW3FcpDxvJbzrBU6Pb2o66FkUzU,6203
190
- litert_torch/generative/export_hf/core/export_lib.py,sha256=4Nr_kaUgtukCMbcYv5MnJmdlrUCau1FeyVR7o-8Y3SU,12005
190
+ litert_torch/generative/export_hf/core/export_lib.py,sha256=BLEmSuaOQpivSWJuw4ei8TGXH_vSu2o9nC43OXpw_lU,12458
191
191
  litert_torch/generative/export_hf/core/exportable_module.py,sha256=XEqsV9M34OP-_vsxH7bnmxSCD6erAPl0a9I9JQM7v6k,8305
192
192
  litert_torch/generative/export_hf/core/litert_lm_builder.py,sha256=f8Q2ifVyt65V-kRL0X9FRpQNKIer0R_Yx2lECZTMGPU,7965
193
193
  litert_torch/generative/export_hf/core/patches.py,sha256=i1fzs0anIFbBH-Q_PwCtp9VKXy64olJKwnGpnJUjkEo,1815
@@ -318,9 +318,9 @@ litert_torch/testing/__init__.py,sha256=AfYP1HwTYSQmupveonEHCDV5dEyshzUgbwUrCUhb
318
318
  litert_torch/testing/export.py,sha256=3dR6oxnrdtX0MfqAfMv233cf3sHA4e0F2TBQotoo8xc,3292
319
319
  litert_torch/testing/model_coverage/__init__.py,sha256=uPXeAhWiD1O0aMDLCX7FTOSNQiea8yOtoIYPCuHEAG4,763
320
320
  litert_torch/testing/model_coverage/model_coverage.py,sha256=EPCI7PbNPb7GV28lo3qQvFdzJwJ_ZDrbCGdpeiBZhVo,4715
321
- litert_torch_nightly-0.9.0.dev20260130.dist-info/licenses/LICENSE,sha256=z8d0m5b2O9McPEK1xHG_dWgUBT6EfBDz6wA0F7xSPTA,11358
322
- litert_torch_nightly-0.9.0.dev20260130.dist-info/METADATA,sha256=nZMdMxwYskn3oPgsfW4biIGlwWhj87dtnuUXw3ewkVo,2463
323
- litert_torch_nightly-0.9.0.dev20260130.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
324
- litert_torch_nightly-0.9.0.dev20260130.dist-info/entry_points.txt,sha256=roYAi9hp0uYrMudMR59hGNF2pz0TSAtqNl4vQLJzxnE,55
325
- litert_torch_nightly-0.9.0.dev20260130.dist-info/top_level.txt,sha256=mGrsl2SYcjQSLBJX4ZXrHnFqHZe6QLRR7uk0tLfzwfM,13
326
- litert_torch_nightly-0.9.0.dev20260130.dist-info/RECORD,,
321
+ litert_torch_nightly-0.9.0.dev20260131.dist-info/licenses/LICENSE,sha256=z8d0m5b2O9McPEK1xHG_dWgUBT6EfBDz6wA0F7xSPTA,11358
322
+ litert_torch_nightly-0.9.0.dev20260131.dist-info/METADATA,sha256=HROruvZl8d-od_o0rkh2JVmwAfh3veOOA6ICPFPNqN0,2463
323
+ litert_torch_nightly-0.9.0.dev20260131.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
324
+ litert_torch_nightly-0.9.0.dev20260131.dist-info/entry_points.txt,sha256=roYAi9hp0uYrMudMR59hGNF2pz0TSAtqNl4vQLJzxnE,55
325
+ litert_torch_nightly-0.9.0.dev20260131.dist-info/top_level.txt,sha256=mGrsl2SYcjQSLBJX4ZXrHnFqHZe6QLRR7uk0tLfzwfM,13
326
+ litert_torch_nightly-0.9.0.dev20260131.dist-info/RECORD,,