fbgemm-gpu-genai-nightly 2025.10.11__cp39-cp39-manylinux_2_28_x86_64.whl → 2025.10.13__cp39-cp39-manylinux_2_28_x86_64.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.

Potentially problematic release.


This version of fbgemm-gpu-genai-nightly might be problematic. Click here for more details.

fbgemm_gpu/asmjit.so CHANGED
Binary file
@@ -6,6 +6,6 @@
6
6
  # This source code is licensed under the BSD-style license found in the
7
7
  # LICENSE file in the root directory of this source tree.
8
8
 
9
- __version__: str = "2025.10.11"
9
+ __version__: str = "2025.10.13"
10
10
  __target__: str = "genai"
11
11
  __variant__: str = "cuda"
fbgemm_gpu/fbgemm.so CHANGED
Binary file
@@ -3971,8 +3971,8 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
3971
3971
  self.step, stats_reporter.report_interval # pyre-ignore
3972
3972
  )
3973
3973
 
3974
- if len(dram_kv_perf_stats) != 24:
3975
- logging.error("dram cache perf stats should have 24 elements")
3974
+ if len(dram_kv_perf_stats) != 36:
3975
+ logging.error("dram cache perf stats should have 36 elements")
3976
3976
  return
3977
3977
 
3978
3978
  dram_read_duration = dram_kv_perf_stats[0]
@@ -4002,6 +4002,19 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
4002
4002
  dram_kv_actual_used_chunk_bytes = dram_kv_perf_stats[21]
4003
4003
  dram_kv_num_rows = dram_kv_perf_stats[22]
4004
4004
  dram_kv_read_counts = dram_kv_perf_stats[23]
4005
+ dram_metadata_write_sharding_total_duration = dram_kv_perf_stats[24]
4006
+ dram_metadata_write_total_duration = dram_kv_perf_stats[25]
4007
+ dram_metadata_write_allocate_avg_duration = dram_kv_perf_stats[26]
4008
+ dram_metadata_write_lookup_cache_avg_duration = dram_kv_perf_stats[27]
4009
+ dram_metadata_write_acquire_lock_avg_duration = dram_kv_perf_stats[28]
4010
+ dram_metadata_write_cache_miss_avg_count = dram_kv_perf_stats[29]
4011
+
4012
+ dram_read_metadata_total_duration = dram_kv_perf_stats[30]
4013
+ dram_read_metadata_sharding_total_duration = dram_kv_perf_stats[31]
4014
+ dram_read_metadata_cache_hit_copy_avg_duration = dram_kv_perf_stats[32]
4015
+ dram_read_metadata_lookup_cache_total_avg_duration = dram_kv_perf_stats[33]
4016
+ dram_read_metadata_acquire_lock_avg_duration = dram_kv_perf_stats[34]
4017
+ dram_read_read_metadata_load_size = dram_kv_perf_stats[35]
4005
4018
 
4006
4019
  stats_reporter.report_duration(
4007
4020
  iteration_step=self.step,
@@ -4168,6 +4181,88 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
4168
4181
  data_bytes=dram_kv_num_rows,
4169
4182
  enable_tb_metrics=True,
4170
4183
  )
4184
+ stats_reporter.report_duration(
4185
+ iteration_step=self.step,
4186
+ event_name="dram_kv.perf.set.dram_eviction_score_write_sharding_total_duration_us",
4187
+ duration_ms=dram_metadata_write_sharding_total_duration,
4188
+ enable_tb_metrics=True,
4189
+ time_unit="us",
4190
+ )
4191
+ stats_reporter.report_duration(
4192
+ iteration_step=self.step,
4193
+ event_name="dram_kv.perf.set.dram_eviction_score_write_total_duration_us",
4194
+ duration_ms=dram_metadata_write_total_duration,
4195
+ enable_tb_metrics=True,
4196
+ time_unit="us",
4197
+ )
4198
+ stats_reporter.report_duration(
4199
+ iteration_step=self.step,
4200
+ event_name="dram_kv.perf.set.dram_eviction_score_write_allocate_avg_duration_us",
4201
+ duration_ms=dram_metadata_write_allocate_avg_duration,
4202
+ enable_tb_metrics=True,
4203
+ time_unit="us",
4204
+ )
4205
+ stats_reporter.report_duration(
4206
+ iteration_step=self.step,
4207
+ event_name="dram_kv.perf.set.dram_eviction_score_write_lookup_cache_avg_duration_us",
4208
+ duration_ms=dram_metadata_write_lookup_cache_avg_duration,
4209
+ enable_tb_metrics=True,
4210
+ time_unit="us",
4211
+ )
4212
+ stats_reporter.report_duration(
4213
+ iteration_step=self.step,
4214
+ event_name="dram_kv.perf.set.dram_eviction_score_write_acquire_lock_avg_duration_us",
4215
+ duration_ms=dram_metadata_write_acquire_lock_avg_duration,
4216
+ enable_tb_metrics=True,
4217
+ time_unit="us",
4218
+ )
4219
+ stats_reporter.report_data_amount(
4220
+ iteration_step=self.step,
4221
+ event_name="dram_kv.perf.set.dram_eviction_score_write_cache_miss_avg_count",
4222
+ data_bytes=dram_metadata_write_cache_miss_avg_count,
4223
+ enable_tb_metrics=True,
4224
+ )
4225
+ stats_reporter.report_duration(
4226
+ iteration_step=self.step,
4227
+ event_name="dram_kv.perf.get.dram_eviction_score_read_total_duration_us",
4228
+ duration_ms=dram_read_metadata_total_duration,
4229
+ enable_tb_metrics=True,
4230
+ time_unit="us",
4231
+ )
4232
+ stats_reporter.report_duration(
4233
+ iteration_step=self.step,
4234
+ event_name="dram_kv.perf.get.dram_eviction_score_read_sharding_total_duration_us",
4235
+ duration_ms=dram_read_metadata_sharding_total_duration,
4236
+ enable_tb_metrics=True,
4237
+ time_unit="us",
4238
+ )
4239
+ stats_reporter.report_duration(
4240
+ iteration_step=self.step,
4241
+ event_name="dram_kv.perf.get.dram_eviction_score_read_cache_hit_copy_avg_duration_us",
4242
+ duration_ms=dram_read_metadata_cache_hit_copy_avg_duration,
4243
+ enable_tb_metrics=True,
4244
+ time_unit="us",
4245
+ )
4246
+ stats_reporter.report_duration(
4247
+ iteration_step=self.step,
4248
+ event_name="dram_kv.perf.get.dram_eviction_score_read_lookup_cache_total_avg_duration_us",
4249
+ duration_ms=dram_read_metadata_lookup_cache_total_avg_duration,
4250
+ enable_tb_metrics=True,
4251
+ time_unit="us",
4252
+ )
4253
+ stats_reporter.report_duration(
4254
+ iteration_step=self.step,
4255
+ event_name="dram_kv.perf.get.dram_eviction_score_read_acquire_lock_avg_duration_us",
4256
+ duration_ms=dram_read_metadata_acquire_lock_avg_duration,
4257
+ enable_tb_metrics=True,
4258
+ time_unit="us",
4259
+ )
4260
+ stats_reporter.report_data_amount(
4261
+ iteration_step=self.step,
4262
+ event_name="dram_kv.perf.get.dram_eviction_score_read_load_size",
4263
+ data_bytes=dram_read_read_metadata_load_size,
4264
+ enable_tb_metrics=True,
4265
+ )
4171
4266
 
4172
4267
  def _recording_to_timer(
4173
4268
  self, timer: Optional[AsyncSeriesTimer], **kwargs: Any
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: fbgemm_gpu_genai_nightly
3
- Version: 2025.10.11
3
+ Version: 2025.10.13
4
4
  Home-page: https://github.com/pytorch/fbgemm
5
5
  Author: FBGEMM Team
6
6
  Author-email: packages@pytorch.org
@@ -1,8 +1,8 @@
1
1
  fbgemm_gpu/__init__.py,sha256=FdQCmpvETH80tlIPP6W8MrOmzLaX9eoGY-fuHtVPbj0,5747
2
- fbgemm_gpu/asmjit.so,sha256=JDql0ZOBynsM53nSLqGG070z1oYke3r68K90AeBP0hY,484232
2
+ fbgemm_gpu/asmjit.so,sha256=RMRQVgIB3mf-bUr2BOMzrgzIv3H9arLfYnJ8CTGZ8RQ,501728
3
3
  fbgemm_gpu/batched_unary_embeddings_ops.py,sha256=GYeJ9pg-Wc9FokXVci_npDsL6UV18-pJXID2xzrJ9O8,2904
4
4
  fbgemm_gpu/enums.py,sha256=37ewGSfO1x7sO31ZkRiqV1yKuklfHXT5qZIxzeeGogo,755
5
- fbgemm_gpu/fbgemm.so,sha256=DEQ1gRPSFaPAUbupt2byc0F7ZpHYo1yagdZ1Se9Gq1A,5807256
5
+ fbgemm_gpu/fbgemm.so,sha256=B9y6MDLC6Ou7Bw_pT07Xfw5g5Q4j2yn9Xsp96QVpgEU,5646712
6
6
  fbgemm_gpu/metrics.py,sha256=TsurFLJf0nJvPDN7urWb4LMQlf5RgdWPTTTDO7S4wtI,5663
7
7
  fbgemm_gpu/permute_pooled_embedding_modules.py,sha256=vOXMYclaGnwSt0St_SOAlAe18kz6WjMyTeHnC9jLhcE,5130
8
8
  fbgemm_gpu/permute_pooled_embedding_modules_split.py,sha256=f3VJvH_kw9Ltd_DXtaf_PJPHmlmEWrQgzQ7MDkhh5Nw,2746
@@ -32,9 +32,9 @@ fbgemm_gpu/docs/merge_pooled_embedding_ops.py,sha256=oJLgSgZQmhsyGLbTmZTxNgQrk65
32
32
  fbgemm_gpu/docs/permute_pooled_embedding_ops.py,sha256=tZUqLVXlk5O6VAKKDA-OEMx2fCu5QPOOeoAPZA9_nLY,4454
33
33
  fbgemm_gpu/docs/quantize_ops.py,sha256=xTtOaVK1P02ymreE_i21YiyYDZCqhoZY9eWp_mEIRlo,1297
34
34
  fbgemm_gpu/docs/sparse_ops.py,sha256=gSLUFdnu8lle_6gLewFkM20wL3ek2jKLvDGMKR6POaY,27292
35
- fbgemm_gpu/docs/version.py,sha256=37Mm_y7ZEW2PuqA4FzAUhP66mtP19dXqT6kcG1kkcHs,317
35
+ fbgemm_gpu/docs/version.py,sha256=C8c36ZYn_Mlg1TksQPQy_4M6KdCMGp-MdnA0uca2nsw,317
36
36
  fbgemm_gpu/experimental/example/__init__.py,sha256=V_XrGMq2oNVMpzwe1srlaTaHeIcZJw5oAGbo3seM_Ks,870
37
- fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=eife1eWK6QFDUtDL-EcAQNvgJyfR86tAGpvDNHsSlpA,232480
37
+ fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=owFe-ilNrqChzBPf7R9BfcY3lpFA6kMuXZn_bKPb2l8,239808
38
38
  fbgemm_gpu/experimental/example/utils.py,sha256=Je__VkMlBMLOhh7NXOocOdvaa2gz9kl9Dkqeu25tpFA,562
39
39
  fbgemm_gpu/experimental/gemm/triton_gemm/__init__.py,sha256=AqHefiOaN_SjP5ew7RYGuKFuSlhedOJL_6f97TtLv7c,566
40
40
  fbgemm_gpu/experimental/gemm/triton_gemm/fp4_quantize.py,sha256=2RjIDSzUXtoFoC2ryp-C-j5H83mbSjPwvsvTrThfrqE,215658
@@ -43,7 +43,7 @@ fbgemm_gpu/experimental/gemm/triton_gemm/grouped_gemm.py,sha256=rbjxTMefjQWgJrWK
43
43
  fbgemm_gpu/experimental/gemm/triton_gemm/matmul_perf_model.py,sha256=SltbY_dsit5e7B8lDIB_VYPrEq0t9kckthj9mQaVNfA,7571
44
44
  fbgemm_gpu/experimental/gemm/triton_gemm/utils.py,sha256=rULXIpVaaRS3GKUZ1RHcWUrUyy0xMVREwS1SFShGgcw,4302
45
45
  fbgemm_gpu/experimental/gen_ai/__init__.py,sha256=qwfuF5E5K4oDiH7RJkpC7zth3kAsG7wv_glCl2A_G2A,1860
46
- fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=twpjHd8M2YkcqqcLaOT7CS7wHII4Gs6ij2n0XvhLBfg,72323352
46
+ fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=Bs1TLo-OcK2urqBHk83UTgHIMLQvFhpU4bgoRkayK74,73016984
47
47
  fbgemm_gpu/experimental/gen_ai/quantize.py,sha256=KAljWSdN-1_c5DWfT-3MDxWLMULK49Yu36t6TmQI9Tw,12599
48
48
  fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/__init__.py,sha256=oExepXpjMOwM43gARZARY0UtR-EX2zqRnSrOaQPy448,1044
49
49
  fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_custom_op.py,sha256=FADVTYzS2u8fA-3iChS5CbtWd0mWF8F3lnXcwr_7vDw,7821
@@ -99,7 +99,7 @@ fbgemm_gpu/tbe/cache/split_embeddings_cache_ops.py,sha256=vZHj7KIe1DoJDy5eft29Xt
99
99
  fbgemm_gpu/tbe/ssd/__init__.py,sha256=wzfMT10cp_dqK2lrebC449hOdexBnizcf_98lA1NyHs,483
100
100
  fbgemm_gpu/tbe/ssd/common.py,sha256=1J8K7sTQswgCYWaVwF-ZdCJj7mNN6O9GI70AaZWzJGE,1044
101
101
  fbgemm_gpu/tbe/ssd/inference.py,sha256=B_uX66ajGA9YKGlFa5TmGWs7b-b1RFigzwxmENZ9Oio,22816
102
- fbgemm_gpu/tbe/ssd/training.py,sha256=iepmavrK6cSXrqWg3TqVXL5gVmSftfHJycesp7I0Nw4,194911
102
+ fbgemm_gpu/tbe/ssd/training.py,sha256=X-u33ZUpxsVUXdQ8UgVtorywAAse5YLrH-ElAG6RnJs,199517
103
103
  fbgemm_gpu/tbe/ssd/utils/__init__.py,sha256=5DgmR2HA6NtmYh2ddkUgpDsZ6a7hF0DPedA1gMpdh18,250
104
104
  fbgemm_gpu/tbe/ssd/utils/partially_materialized_tensor.py,sha256=SFg2-29b-i49LWm-FlaWUkTz2XzXbicYi_AzVj4jKNE,7601
105
105
  fbgemm_gpu/tbe/stats/__init__.py,sha256=on29iDtq7cVNh90JR9aeFNG-K9DDoYq0JryzoplL49I,322
@@ -121,7 +121,7 @@ fbgemm_gpu/utils/loader.py,sha256=1hCEhNvkflniH46fGcrguLeP1z-6uyOu2QFwqKU5CIM,99
121
121
  fbgemm_gpu/utils/torch_library.py,sha256=ywsAHjbuwesj50LjEu99WkAH17FlaVgePZ9OmFg6YE4,4193
122
122
  list_versions/__init__.py,sha256=UmTeqCk-UJWFtlZQWvZao3xvui2w9E3X_JdOXVjRaNw,315
123
123
  list_versions/cli_run.py,sha256=CChZoXQ-tiKaWboXAYlPVJ5w8K5zAKiKcncA087I1sc,4508
124
- fbgemm_gpu_genai_nightly-2025.10.11.dist-info/METADATA,sha256=6M2BPNwybS1j-b2XjITa24q7t8uq-mIq-MEafDcYh70,2656
125
- fbgemm_gpu_genai_nightly-2025.10.11.dist-info/WHEEL,sha256=LLKFkWvXQZlw1kR7spdgzB2PptbzNM_HUUPM3Q0J_XE,106
126
- fbgemm_gpu_genai_nightly-2025.10.11.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
127
- fbgemm_gpu_genai_nightly-2025.10.11.dist-info/RECORD,,
124
+ fbgemm_gpu_genai_nightly-2025.10.13.dist-info/METADATA,sha256=zYHDD18ERZu517cxjQmIwbCua0gVPrlcjhkFHNrv06A,2656
125
+ fbgemm_gpu_genai_nightly-2025.10.13.dist-info/WHEEL,sha256=LLKFkWvXQZlw1kR7spdgzB2PptbzNM_HUUPM3Q0J_XE,106
126
+ fbgemm_gpu_genai_nightly-2025.10.13.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
127
+ fbgemm_gpu_genai_nightly-2025.10.13.dist-info/RECORD,,