fbgemm-gpu-genai-nightly 2025.10.28__cp312-cp312-manylinux_2_28_x86_64.whl → 2025.10.31__cp312-cp312-manylinux_2_28_x86_64.whl

fbgemm_gpu/docs/target.genai.json.py

@@ -1,6 +1,6 @@
 
 {
-    "version": "2025.10.28",
+    "version": "2025.10.31",
     "target": "genai",
     "variant": "cuda"
 }

fbgemm_gpu/split_table_batched_embeddings_ops_common.py

@@ -240,6 +240,19 @@ class KVZCHParams(NamedTuple):
         ), "backend_return_whole_row can only be enabled when enable_optimizer_offloading is enabled"
 
 
+class KVZCHEvictionTBEConfig(NamedTuple):
+    # Eviction trigger model for kvzch table: 0: disabled, 1: iteration, 2: mem_util, 3: manual, 4: id count, 5: free_mem
+    kvzch_eviction_trigger_mode: int = 2  # mem_util
+    # Minimum free memory (in GB) required before triggering eviction when using free_mem trigger mode.
+    eviction_free_mem_threshold_gb: int = 200  # 200GB
+    # Number of batches between checks for free memory threshold when using free_mem trigger mode.
+    eviction_free_mem_check_interval_batch: int = 1000
+    # The width of each feature score bucket used for threshold calculation in feature score-based eviction.
+    threshold_calculation_bucket_stride: float = 0.2
+    # Total number of feature score buckets used for threshold calculation in feature score-based eviction.
+    threshold_calculation_bucket_num: Optional[int] = 1000000  # 1M
+
+
 class BackendType(enum.IntEnum):
     SSD = 0
     DRAM = 1

fbgemm_gpu/split_table_batched_embeddings_ops_training.py

@@ -1556,7 +1556,7 @@ class SplitTableBatchedEmbeddingBagsCodegen(nn.Module):
         table_name_set = set(table_names)
         if len(table_name_set) == 1:
             return next(iter(table_name_set))
-        return f"<{len(table_name_set)} tables>"
+        return f"<{len(table_name_set)} tables>: {table_name_set}"
 
     @staticmethod
     def get_prefetch_passes(

fbgemm_gpu/tbe/ssd/training.py

@@ -2089,7 +2089,7 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
                         torch.tensor(
                             [weights.shape[0]], device="cpu", dtype=torch.long
                         ),
-                        weights.cpu().view(torch.float32).view(-1, 2),
+                        weights.cpu(),
                     )
 
             # Generate row addresses (pointing to either L1 or the current

{fbgemm_gpu_genai_nightly-2025.10.28.dist-info → fbgemm_gpu_genai_nightly-2025.10.31.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fbgemm_gpu_genai_nightly
-Version: 2025.10.28
+Version: 2025.10.31
 Home-page: https://github.com/pytorch/fbgemm
 Author: FBGEMM Team
 Author-email: packages@pytorch.org

{fbgemm_gpu_genai_nightly-2025.10.28.dist-info → fbgemm_gpu_genai_nightly-2025.10.31.dist-info}/RECORD

@@ -1,8 +1,8 @@
 fbgemm_gpu/__init__.py,sha256=A3DuseilQ-sEtBpeZsG0LOqN5Cl3e5DHI_YgCZEMhnE,6283
-fbgemm_gpu/asmjit.so,sha256=yDq47YobRro7Tvd4IaPNyQUf1YaA8iLyfcwnUdh0Coo,484232
+fbgemm_gpu/asmjit.so,sha256=tp-5cN7HUYo7cjvR_kl_vfPBSEv78-IQxdvHN-nXFAM,501728
 fbgemm_gpu/batched_unary_embeddings_ops.py,sha256=GYeJ9pg-Wc9FokXVci_npDsL6UV18-pJXID2xzrJ9O8,2904
 fbgemm_gpu/enums.py,sha256=37ewGSfO1x7sO31ZkRiqV1yKuklfHXT5qZIxzeeGogo,755
-fbgemm_gpu/fbgemm.so,sha256=DEQ1gRPSFaPAUbupt2byc0F7ZpHYo1yagdZ1Se9Gq1A,5807256
+fbgemm_gpu/fbgemm.so,sha256=B9y6MDLC6Ou7Bw_pT07Xfw5g5Q4j2yn9Xsp96QVpgEU,5646712
 fbgemm_gpu/metrics.py,sha256=TsurFLJf0nJvPDN7urWb4LMQlf5RgdWPTTTDO7S4wtI,5663
 fbgemm_gpu/permute_pooled_embedding_modules.py,sha256=vOXMYclaGnwSt0St_SOAlAe18kz6WjMyTeHnC9jLhcE,5130
 fbgemm_gpu/permute_pooled_embedding_modules_split.py,sha256=f3VJvH_kw9Ltd_DXtaf_PJPHmlmEWrQgzQ7MDkhh5Nw,2746
@@ -15,9 +15,9 @@ fbgemm_gpu/split_embedding_inference_converter.py,sha256=AghGW22MgMsdHzdwdPMPYDj
 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=76ME0692CC691xpjiOsY3Xxy-LD_XKs8w9vq1gcm9tM,16440
+fbgemm_gpu/split_table_batched_embeddings_ops_common.py,sha256=Y7q0HbxTehMAYaow-A0MOAjsgyyB4PB31VGP4S_MDKc,17310
 fbgemm_gpu/split_table_batched_embeddings_ops_inference.py,sha256=dGC85xjQiRUrequBibSf9oMAVHT5Q49zsVo2zW4n_88,81679
-fbgemm_gpu/split_table_batched_embeddings_ops_training.py,sha256=f0sXfvkE0Wx0Rd3qTT4XmCbBK0wYgWGzhPncZEv-p48,180420
+fbgemm_gpu/split_table_batched_embeddings_ops_training.py,sha256=sM4xXXU6p_clbPIRLO7UI5fnIcan8dt2FhL6KURZTQw,180438
 fbgemm_gpu/split_table_batched_embeddings_ops_training_common.py,sha256=e3O9ElaWBGvG7TdT3Ok_8cB06jhskXuyCQ0t40dzsEY,5449
 fbgemm_gpu/ssd_split_table_batched_embeddings_ops.py,sha256=7qGkO8FARku38mFYl4Bc4qL8dS1wrfyorS9l1m5ZAVA,718
 fbgemm_gpu/tbe_input_multiplexer.py,sha256=TQjwkJ2JkOaQsMYuRdk9RbNa9759EPEtx8bYclChtZY,3063
@@ -32,9 +32,9 @@ 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=gSLUFdnu8lle_6gLewFkM20wL3ek2jKLvDGMKR6POaY,27292
-fbgemm_gpu/docs/target.genai.json.py,sha256=8qUPmreVJ1_8egLGQELc4MkLwuD8aXPlYC2OfOCPJ3g,79
+fbgemm_gpu/docs/target.genai.json.py,sha256=lnTGpquMN5G34NLn02Vfksm8F1T2UsoJ7fOnDUg1f9w,79
 fbgemm_gpu/experimental/example/__init__.py,sha256=OvJHZgWnycL1gWKyCXFJCTKuys3KAqx4iadjx3R-tBQ,723
-fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=K70etfWeSleFOhfxXvSmpZMYBn_xmpvSxgdcGenvaKo,232488
+fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=d5yEUZZmFfjLNZmSi2vd-JlxouPL7OP_kNE_p3zbcvo,243904
 fbgemm_gpu/experimental/example/utils.py,sha256=Je__VkMlBMLOhh7NXOocOdvaa2gz9kl9Dkqeu25tpFA,562
 fbgemm_gpu/experimental/gemm/triton_gemm/__init__.py,sha256=1CqUfzlYyXTvU-BNaUq4RZpLV-2lKAVCAHeJzSIZFWw,419
 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
 fbgemm_gpu/experimental/gemm/triton_gemm/matmul_perf_model.py,sha256=SltbY_dsit5e7B8lDIB_VYPrEq0t9kckthj9mQaVNfA,7571
 fbgemm_gpu/experimental/gemm/triton_gemm/utils.py,sha256=rULXIpVaaRS3GKUZ1RHcWUrUyy0xMVREwS1SFShGgcw,4302
 fbgemm_gpu/experimental/gen_ai/__init__.py,sha256=r3NlNCXuIh0pfKwKU5v14y6AZkpoIkKWbtzxSprgeKA,1713
-fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=f1pv2D7fQM-XxvuDGUtLPN2ZTTbQIMNN7GE0AYaHF4g,74306608
+fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=pcOjuee-OJQJmo4z9Iw0QXYcSEh9RunalMpfZhwD-pQ,74888976
 fbgemm_gpu/experimental/gen_ai/quantize.py,sha256=KAljWSdN-1_c5DWfT-3MDxWLMULK49Yu36t6TmQI9Tw,12599
 fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/__init__.py,sha256=ntFgFs0foi6NQx8eqs5I3fCjzKSI0spXfEWiMhlcT00,897
 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
 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=B_uX66ajGA9YKGlFa5TmGWs7b-b1RFigzwxmENZ9Oio,22816
-fbgemm_gpu/tbe/ssd/training.py,sha256=t_8AfKlajVNFAnn3hPST2e82P0dPxzXAsioxMBJ8jrY,199948
+fbgemm_gpu/tbe/ssd/training.py,sha256=mKi5nC1p6p0ZFG0XBWDqzNDkxE-v2PzFnVI0SY7w9v4,199916
 fbgemm_gpu/tbe/ssd/utils/__init__.py,sha256=5DgmR2HA6NtmYh2ddkUgpDsZ6a7hF0DPedA1gMpdh18,250
 fbgemm_gpu/tbe/ssd/utils/partially_materialized_tensor.py,sha256=SFg2-29b-i49LWm-FlaWUkTz2XzXbicYi_AzVj4jKNE,7601
 fbgemm_gpu/tbe/stats/__init__.py,sha256=on29iDtq7cVNh90JR9aeFNG-K9DDoYq0JryzoplL49I,322
@@ -121,7 +121,7 @@ fbgemm_gpu/utils/loader.py,sha256=1hCEhNvkflniH46fGcrguLeP1z-6uyOu2QFwqKU5CIM,99
 fbgemm_gpu/utils/torch_library.py,sha256=ywsAHjbuwesj50LjEu99WkAH17FlaVgePZ9OmFg6YE4,4193
 list_versions/__init__.py,sha256=UmTeqCk-UJWFtlZQWvZao3xvui2w9E3X_JdOXVjRaNw,315
 list_versions/cli_run.py,sha256=CChZoXQ-tiKaWboXAYlPVJ5w8K5zAKiKcncA087I1sc,4508
-fbgemm_gpu_genai_nightly-2025.10.28.dist-info/METADATA,sha256=06TbYlUoKnFZX6E1ZpX2rcQaq4tSOj0kJxNOSQ2yO8M,2656
-fbgemm_gpu_genai_nightly-2025.10.28.dist-info/WHEEL,sha256=vUT1hK8fT5m5CAs5kDyQ_ABrvCmtd0TCp5-4vN9tR5A,108
-fbgemm_gpu_genai_nightly-2025.10.28.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
-fbgemm_gpu_genai_nightly-2025.10.28.dist-info/RECORD,,
+fbgemm_gpu_genai_nightly-2025.10.31.dist-info/METADATA,sha256=vJe0nced-mx3C1Ti6JrR_9TXc-GSQHsDteKtvodR8jg,2656
+fbgemm_gpu_genai_nightly-2025.10.31.dist-info/WHEEL,sha256=vUT1hK8fT5m5CAs5kDyQ_ABrvCmtd0TCp5-4vN9tR5A,108
+fbgemm_gpu_genai_nightly-2025.10.31.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
+fbgemm_gpu_genai_nightly-2025.10.31.dist-info/RECORD,,