fbgemm-gpu-hstu-nightly 2025.6.12__cp310-cp310-manylinux_2_28_x86_64.whl → 2025.6.14__cp310-cp310-manylinux_2_28_x86_64.whl

fbgemm_gpu/docs/version.py

@@ -6,6 +6,6 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-__version__: str = "2025.6.12"
+__version__: str = "2025.6.14"
 __target__: str = "hstu"
 __variant__: str = "cuda"

fbgemm_gpu/sparse_ops.py

@@ -420,6 +420,7 @@ def int_nbit_split_embedding_codegen_lookup_function_meta(
     kINT8QparamsBytes = 8
 
     if pooling_mode == PoolingMode.NONE:
+        kINT8QparamsBytes = 4
         D = max(
             [
                 max_int2_D,
@@ -435,7 +436,7 @@ def int_nbit_split_embedding_codegen_lookup_function_meta(
         torch._check(D > 0)
         adjusted_D = D
         if SparseType.from_int(output_dtype_int) == SparseType.INT8:
-            adjusted_D += T * kINT8QparamsBytes
+            adjusted_D += kINT8QparamsBytes
         output = dev_weights.new_empty([total_L, adjusted_D], dtype=output_dtype)
         return output
 

fbgemm_gpu/split_table_batched_embeddings_ops_common.py

@@ -33,6 +33,17 @@ class EmbeddingLocation(enum.IntEnum):
     HOST = 3
     MTIA = 4
 
+    @classmethod
+    # pyre-ignore[3]
+    def str_values(cls):
+        return [
+            "device",
+            "managed",
+            "managed_caching",
+            "host",
+            "mtia",
+        ]
+
     @classmethod
     # pyre-ignore[3]
     def from_str(cls, key: str):

fbgemm_gpu/tbe/bench/embedding_ops_common_config.py

@@ -99,9 +99,7 @@ class EmbeddingOpsCommonConfigLoader:
             click.option(
                 "--emb-location",
                 default="device",
-                type=click.Choice(
-                    ["device", "managed", "managed_caching"], case_sensitive=False
-                ),
+                type=click.Choice(EmbeddingLocation.str_values(), case_sensitive=False),
                 help="Memory location of the embeddings",
             ),
             click.option(

fbgemm_gpu/tbe/ssd/training.py

@@ -2804,7 +2804,12 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
         """
         Create a rocksdb hard link snapshot to provide cross procs access to the underlying data
         """
-        self.ssd_db.create_rocksdb_hard_link_snapshot(self.step)
+        if self.backend_type == BackendType.SSD:
+            self.ssd_db.create_rocksdb_hard_link_snapshot(self.step)
+        else:
+            logging.warning(
+                "create_rocksdb_hard_link_snapshot is only supported for SSD backend"
+            )
 
     def prepare_inputs(
         self,

fbgemm_gpu/tbe/ssd/utils/partially_materialized_tensor.py

@@ -9,7 +9,7 @@
 from __future__ import annotations
 
 import functools
-from typing import Optional, Union
+from typing import List, Optional, Union
 
 import torch
 
@@ -249,6 +249,9 @@ class PartiallyMaterializedTensor:
 
         return torch.equal(tensor1.full_tensor(), tensor2.full_tensor())
 
+    def get_kvtensor_serializable_metadata(self) -> List[str]:
+        return self._wrapped.get_kvtensor_serializable_metadata()
+
     def __hash__(self):
         return id(self)
 

{fbgemm_gpu_hstu_nightly-2025.6.12.dist-info → fbgemm_gpu_hstu_nightly-2025.6.14.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fbgemm_gpu_hstu_nightly
-Version: 2025.6.12
+Version: 2025.6.14
 Home-page: https://github.com/pytorch/fbgemm
 Author: FBGEMM Team
 Author-email: packages@pytorch.org

{fbgemm_gpu_hstu_nightly-2025.6.12.dist-info → fbgemm_gpu_hstu_nightly-2025.6.14.dist-info}/RECORD

@@ -9,13 +9,13 @@ fbgemm_gpu/permute_pooled_embedding_modules_split.py,sha256=cUrEbRIvLFW_3Zmh07Qk
 fbgemm_gpu/quantize_comm.py,sha256=YUzk8F1MZckbkseEoDJ4rOKiglGCGtrPdR1IKGD6Mk4,11177
 fbgemm_gpu/quantize_utils.py,sha256=hb8G_1xzRwYmwFp8VJrmoaolNxCwqcFwkwnyStk1C0w,7394
 fbgemm_gpu/runtime_monitor.py,sha256=HM_0cxMO7uuAq8sCiv2lmGgp1jKGzba2qhuUcGyRMog,7425
-fbgemm_gpu/sparse_ops.py,sha256=84x_hobQofY29Vzlz0eJxe126Ba-O5oSEQREMpRgOtE,47168
+fbgemm_gpu/sparse_ops.py,sha256=xNoRMp6QNMz8Lq_5LE9IprQtUip3gkTVbyPgJ1AiWSI,47194
 fbgemm_gpu/split_embedding_configs.py,sha256=DcZ7SV4AmhlN9QPDaJBqzigR-c1zM_bZI3Fh4PYuab4,7266
 fbgemm_gpu/split_embedding_inference_converter.py,sha256=ilVVowkTiY0WDpOYorj917Tqsez4KWNBdTXuz2bWbp8,7063
 fbgemm_gpu/split_embedding_optimizer_ops.py,sha256=wXuGazClBMk62yL_r9udUIKaPgQP7SlkSb5ugB75wrQ,711
 fbgemm_gpu/split_embedding_utils.py,sha256=Gb40ZKeATxIKEKI3aVQMgDDBanNpKMc53Z43mnzdR_I,851
 fbgemm_gpu/split_table_batched_embeddings_ops.py,sha256=_MIp6uHYHLn4GxGdrGsfddfSsZ2Z9mjsYIrih3ncI1I,2339
-fbgemm_gpu/split_table_batched_embeddings_ops_common.py,sha256=jRCuFRUPchgqL9Z9RTEIjZGl9R6YJe1RlU_Su7AgCMw,8030
+fbgemm_gpu/split_table_batched_embeddings_ops_common.py,sha256=qglNRKKuHkrKiTw90ACjZpMzcjHKXKV7ME3a8QHfQt4,8237
 fbgemm_gpu/split_table_batched_embeddings_ops_inference.py,sha256=bUDWa6IR0vGLDThgB3nmD1yfYa8_HD34B0dtLnd7thw,81692
 fbgemm_gpu/split_table_batched_embeddings_ops_training.py,sha256=YCLPSW9CXrRwMN5KEU6x0ESbutdhzKTaNOO8oN5kX7I,163875
 fbgemm_gpu/split_table_batched_embeddings_ops_training_common.py,sha256=ktC10-nakOBpcmJNCOGQsxuBCP8XTwXJ2WeEgIg91tc,5455
@@ -32,10 +32,10 @@ fbgemm_gpu/docs/merge_pooled_embedding_ops.py,sha256=oJLgSgZQmhsyGLbTmZTxNgQrk65
 fbgemm_gpu/docs/permute_pooled_embedding_ops.py,sha256=tZUqLVXlk5O6VAKKDA-OEMx2fCu5QPOOeoAPZA9_nLY,4454
 fbgemm_gpu/docs/quantize_ops.py,sha256=xTtOaVK1P02ymreE_i21YiyYDZCqhoZY9eWp_mEIRlo,1297
 fbgemm_gpu/docs/sparse_ops.py,sha256=NTcTm0q9h8W2B8PKPoic2fHsAaCbCYunSa_EYK0LtHQ,21382
-fbgemm_gpu/docs/version.py,sha256=XOS_m9ZE4Pr3zwz272OUxVj3s034QJ5xDrpkMLk-x5s,315
+fbgemm_gpu/docs/version.py,sha256=LK5E-TYZjZc9TFmB0M0k8KZYtBxvhQwYFL31OtDqIOU,315
 fbgemm_gpu/experimental/hstu/__init__.py,sha256=KNisP6qDMwgjgxkGlqUZRNjJ_8o8R-cTmm3HxF7pSqI,1564
 fbgemm_gpu/experimental/hstu/cuda_hstu_attention.py,sha256=5425GRjJuzpXQC-TowgQOCFjZmOwv_EK0lKbURhHBTQ,9920
-fbgemm_gpu/experimental/hstu/fbgemm_gpu_experimental_hstu.so,sha256=Ykro0hs9HwqkRXJrWPDjNPn4vzexnNP6ehIKyzC3naY,352287576
+fbgemm_gpu/experimental/hstu/fbgemm_gpu_experimental_hstu.so,sha256=Byu1Kwf8AcaUaY31IKYMUnfI6ncK0ZhDYBqxOhopmow,352287576
 fbgemm_gpu/quantize/__init__.py,sha256=pftciXHE7csekDFkl7Ui1AWglVMMnSrOO04mREnUdb0,921
 fbgemm_gpu/quantize/quantize_ops.py,sha256=25AIOv9n2UoxamMUaI6EK1Ur4gSHxbZIReHBtgOjjCs,2228
 fbgemm_gpu/sll/__init__.py,sha256=rgXh35-OFUE54E9gGBq3NGxouGvgMv2ccY2bWUTxONY,4191
@@ -61,7 +61,7 @@ fbgemm_gpu/tbe/bench/__init__.py,sha256=Uhk3IzAs9f_2kVYcMuFP3KiwwkqqYxaZFivPaBIN
 fbgemm_gpu/tbe/bench/bench_config.py,sha256=GYxVotMi-JK7AEm96sOfB2uOlmRtXpOTr-tkPEB1Q48,4241
 fbgemm_gpu/tbe/bench/bench_runs.py,sha256=hmoldOLh2UsjPAeb0zdfm5aTbs_0iEnlCxsXCHuzAMg,18343
 fbgemm_gpu/tbe/bench/eeg_cli.py,sha256=T8Wa1PeRyFZ0Ge-SErHQEYDY8LvHVoCV_qQlE_6kER0,3600
-fbgemm_gpu/tbe/bench/embedding_ops_common_config.py,sha256=yBk0zaZJ5b3ffPG91SoqD088shrLYdyCFGfn6DsXK9A,5021
+fbgemm_gpu/tbe/bench/embedding_ops_common_config.py,sha256=mdG3JZwgclp6DiVwQSKl5jrirLSId4OuM64knj9TkEk,4973
 fbgemm_gpu/tbe/bench/eval_compression.py,sha256=bINVERk42VJDSdenQHKWApmRMrW8rhkevOgE0hDR-S8,3499
 fbgemm_gpu/tbe/bench/reporter.py,sha256=ZK5RFolUmZEcsEaife270_iOdXAQD5EjTUkuxctnAbY,804
 fbgemm_gpu/tbe/bench/tbe_data_config.py,sha256=uvepBrbzERALBB-RPZVGFra4a8ALCqsOe9X6iWpqAyU,9413
@@ -73,9 +73,9 @@ fbgemm_gpu/tbe/cache/split_embeddings_cache_ops.py,sha256=mQkCl0xN8xUu5bjEWcOOFN
 fbgemm_gpu/tbe/ssd/__init__.py,sha256=wzfMT10cp_dqK2lrebC449hOdexBnizcf_98lA1NyHs,483
 fbgemm_gpu/tbe/ssd/common.py,sha256=1J8K7sTQswgCYWaVwF-ZdCJj7mNN6O9GI70AaZWzJGE,1044
 fbgemm_gpu/tbe/ssd/inference.py,sha256=DTjwj3f6JaUMcecWoRNkZpRgXDJ-eE3grtixYwKb5DI,22829
-fbgemm_gpu/tbe/ssd/training.py,sha256=gCvMY8fGP2JNosMyzW7uiQUuUsDIdUULB_PG51wOpRk,131519
+fbgemm_gpu/tbe/ssd/training.py,sha256=Dx-rJqjrD1A4U4MEVaP3OJl3CZz0VRSTWcukx5557Jw,131715
 fbgemm_gpu/tbe/ssd/utils/__init__.py,sha256=5DgmR2HA6NtmYh2ddkUgpDsZ6a7hF0DPedA1gMpdh18,250
-fbgemm_gpu/tbe/ssd/utils/partially_materialized_tensor.py,sha256=V18ZQxdYJNSi6qLqhFaxXj3IxpcPgan3GsoXrosoZ1Q,7510
+fbgemm_gpu/tbe/ssd/utils/partially_materialized_tensor.py,sha256=uwwEdUiaVlnWZ_rQax2z28VYROfivdMqIdWLy8IZ6cE,7646
 fbgemm_gpu/tbe/stats/__init__.py,sha256=on29iDtq7cVNh90JR9aeFNG-K9DDoYq0JryzoplL49I,322
 fbgemm_gpu/tbe/stats/bench_params_reporter.py,sha256=7XIWVObJOxSVUG73xsd_lVSuCFUQkMEGSWW--BoyCH0,7358
 fbgemm_gpu/tbe/utils/__init__.py,sha256=rlXFm-kTByFZO4SS5C5zMzANRiQmM1NT__eWBayncYg,549
@@ -93,7 +93,7 @@ fbgemm_gpu/utils/__init__.py,sha256=JQQNdcTTaEU6ptK-OW-ZQBwTFxEZZpWOtBXWwEZm39o,
 fbgemm_gpu/utils/filestore.py,sha256=Zshw1dA03m9aHMMAtETdq4bgOLocyLhzlkAUoG8VkdM,4743
 fbgemm_gpu/utils/loader.py,sha256=1hCEhNvkflniH46fGcrguLeP1z-6uyOu2QFwqKU5CIM,990
 fbgemm_gpu/utils/torch_library.py,sha256=dQcHv1qgpu5QYlJjxjd6oeHjtxnmmXzx3PL6vjCmxL4,4199
-fbgemm_gpu_hstu_nightly-2025.6.12.dist-info/METADATA,sha256=ayz07_xnYRnh1FYOD4_6vcF3Pn4Ngw8hmoQhOBEfOz8,2794
-fbgemm_gpu_hstu_nightly-2025.6.12.dist-info/WHEEL,sha256=k9CVMKlTmOLLXq_OyiiJFbPd6UKfogV4yIUezgPmplE,108
-fbgemm_gpu_hstu_nightly-2025.6.12.dist-info/top_level.txt,sha256=2tlbTWLkPjhqvLF_6BbqKzkcPluSE-oPRVjI8axK76I,11
-fbgemm_gpu_hstu_nightly-2025.6.12.dist-info/RECORD,,
+fbgemm_gpu_hstu_nightly-2025.6.14.dist-info/METADATA,sha256=8j-rn3gGFwyI1nr2yksN6x6fTPAhHZ_oXvyn5keeM1k,2794
+fbgemm_gpu_hstu_nightly-2025.6.14.dist-info/WHEEL,sha256=k9CVMKlTmOLLXq_OyiiJFbPd6UKfogV4yIUezgPmplE,108
+fbgemm_gpu_hstu_nightly-2025.6.14.dist-info/top_level.txt,sha256=2tlbTWLkPjhqvLF_6BbqKzkcPluSE-oPRVjI8axK76I,11
+fbgemm_gpu_hstu_nightly-2025.6.14.dist-info/RECORD,,