fbgemm-gpu-genai-nightly 2025.9.29__cp311-cp311-manylinux_2_28_x86_64.whl → 2025.10.2__cp311-cp311-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 = "2025.9.29"
+__version__: str = "2025.10.2"
 __target__: str = "genai"
 __variant__: str = "cuda"

fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py

@@ -3814,7 +3814,6 @@ def get_full_non_persistent_tuning_space():
                                                     num_stages=num_stages,
                                                 )
                                             )
-    logger.info(f"all configs #: {len(configs)}")
     return configs
 
 
@@ -3839,6 +3838,8 @@ _MATMUL_CONFIG_TUPLES_PINGPONG_4K_8K_16K = [
     (128, 64, 64, 4, 1, 0, 16, 2, 4, 2),
     (128, 64, 64, 1, 1, 0, 16, 2, 4, 2),
     (256, 128, 128, 1, 1, 2, 16, 1, 8, 2),
+    (128, 256, 128, 2, 1, 2, 16, 2, 4, 1),
+    (256, 128, 64, 2, 1, 2, 16, 1, 4, 2),
 ]
 
 

fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_custom_op.py

@@ -13,13 +13,13 @@ from torch.library import register_fake
 torch.library.define(
     "blackwell_fmha::fmha_fwd",
     "(Tensor q, Tensor k, Tensor v, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, int? max_seq_len_q, int? max_seq_len_k, float? softmax_scale, bool? causal, Tensor? seqlen_kv) -> (Tensor, Tensor)",
-    tags=[torch.Tag.pt2_compliant_tag],
+    tags=torch.Tag.pt2_compliant_tag,
 )
 
 torch.library.define(
     "blackwell_fmha::fmha_bwd",
     "(Tensor dout, Tensor q, Tensor k, Tensor v, Tensor out, Tensor softmax_lse, Tensor? cu_seqlens_q, Tensor? cu_seqlens_k, int? max_seq_len_q, int? max_seq_len_k, bool? causal) -> (Tensor, Tensor, Tensor)",
-    tags=[torch.Tag.pt2_compliant_tag],
+    tags=torch.Tag.pt2_compliant_tag,
 )
 
 

fbgemm_gpu/tbe/ssd/training.py

@@ -3971,8 +3971,8 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
             self.step, stats_reporter.report_interval  # pyre-ignore
         )
 
-        if len(dram_kv_perf_stats) != 23:
-            logging.error("dram cache perf stats should have 23 elements")
+        if len(dram_kv_perf_stats) != 24:
+            logging.error("dram cache perf stats should have 24 elements")
             return
 
         dram_read_duration = dram_kv_perf_stats[0]
@@ -4001,6 +4001,7 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
         dram_kv_allocated_bytes = dram_kv_perf_stats[20]
         dram_kv_actual_used_chunk_bytes = dram_kv_perf_stats[21]
         dram_kv_num_rows = dram_kv_perf_stats[22]
+        dram_kv_read_counts = dram_kv_perf_stats[23]
 
         stats_reporter.report_duration(
             iteration_step=self.step,
@@ -4142,6 +4143,13 @@ class SSDTableBatchedEmbeddingBags(nn.Module):
             enable_tb_metrics=True,
         )
 
+        stats_reporter.report_data_amount(
+            iteration_step=self.step,
+            event_name="dram_kv.perf.get.dram_kv_read_counts",
+            data_bytes=dram_kv_read_counts,
+            enable_tb_metrics=True,
+        )
+
         stats_reporter.report_data_amount(
             iteration_step=self.step,
             event_name=self.dram_kv_allocated_bytes_stats_name,

{fbgemm_gpu_genai_nightly-2025.9.29.dist-info → fbgemm_gpu_genai_nightly-2025.10.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fbgemm_gpu_genai_nightly
-Version: 2025.9.29
+Version: 2025.10.2
 Home-page: https://github.com/pytorch/fbgemm
 Author: FBGEMM Team
 Author-email: packages@pytorch.org

{fbgemm_gpu_genai_nightly-2025.9.29.dist-info → fbgemm_gpu_genai_nightly-2025.10.2.dist-info}/RECORD

@@ -1,8 +1,8 @@
 fbgemm_gpu/__init__.py,sha256=FdQCmpvETH80tlIPP6W8MrOmzLaX9eoGY-fuHtVPbj0,5747
-fbgemm_gpu/asmjit.so,sha256=RxTYI8zY4PpIBRpSKT_-U7bRIVeTRohdtRFUmLNU1tQ,501728
+fbgemm_gpu/asmjit.so,sha256=tf4jzD7HrM0qcrQl_7Q_y3FJ62duD7b3tKkh5TAN7k8,484232
 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=P-80NThzhyQWN9WMb2kYfl04sAgPCehbdArfnktJaqw,5634424
+fbgemm_gpu/fbgemm.so,sha256=HdsyKYHtVNz5ZNpTcSI4SDuYAGks1hYeVz45ZPjuJts,5790800
 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
@@ -32,21 +32,21 @@ 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/version.py,sha256=F5j2fIxoKrCau1X_4joWddM6S3n3XwdUZvZg5Gc_F48,316
+fbgemm_gpu/docs/version.py,sha256=U9HFTyqt_827sXJZ7N9Dik7e18vj0x7B38Go9HoScG4,316
 fbgemm_gpu/experimental/example/__init__.py,sha256=V_XrGMq2oNVMpzwe1srlaTaHeIcZJw5oAGbo3seM_Ks,870
-fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=-VrIV7q3ZUbZSwgISxDhp5ch8YiOkC6ftvwwQ9UNcb8,243904
+fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=1dZS9lk6C74qym6nRYRikoQlfcRelfkL4i8v2stWylY,232488
 fbgemm_gpu/experimental/example/utils.py,sha256=Je__VkMlBMLOhh7NXOocOdvaa2gz9kl9Dkqeu25tpFA,562
 fbgemm_gpu/experimental/gemm/triton_gemm/__init__.py,sha256=AqHefiOaN_SjP5ew7RYGuKFuSlhedOJL_6f97TtLv7c,566
 fbgemm_gpu/experimental/gemm/triton_gemm/fp4_quantize.py,sha256=I2xf2DlU27KA9s0256tkGLhdOoImUv7i7oHc8bz5Y2M,211841
-fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py,sha256=iB13OssIfH1ShYn-L-Kwd0hFrvkInFnK9YXw248EdMA,152719
+fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py,sha256=5m4SdgUsf2rM_Vul8czgRn_5oVnyi-52TmeidXh05hg,152754
 fbgemm_gpu/experimental/gemm/triton_gemm/grouped_gemm.py,sha256=rbjxTMefjQWgJrWK_bYFtBklJigFwv4awPeVexkkiIA,44511
 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=qwfuF5E5K4oDiH7RJkpC7zth3kAsG7wv_glCl2A_G2A,1860
-fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=2Xp9jIeo84oH-Y0sjSU5KC5eAjU_52lZAap03P3E33c,78714952
+fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=crMx6rSmrjg53kw2gLRXCksM4G-vnqDzxEmHFuRuPYk,78050824
 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=oExepXpjMOwM43gARZARY0UtR-EX2zqRnSrOaQPy448,1044
-fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_custom_op.py,sha256=D90VgPOxnx1NpnDnajIv_L7AHq4rrmFIch0iV2elAVU,7825
+fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_custom_op.py,sha256=FADVTYzS2u8fA-3iChS5CbtWd0mWF8F3lnXcwr_7vDw,7821
 fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_interface.py,sha256=sWk5888_e7Qhfik0X1uZ1VsEUmY5uidxHcqtpoH42Is,9406
 fbgemm_gpu/experimental/gen_ai/bench/__init__.py,sha256=GvCUF6o7wCR3XSWingWKxn_Y3_F2GhZtOIRAB3pfqK0,551
 fbgemm_gpu/experimental/gen_ai/bench/comm_bench.py,sha256=ApEyJOf_rdIo8V_EgvhZXBGNov8ITC_dnB95v8szulI,8515
@@ -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=dnUHnl9FzPneXrWUCA_HKvLI5QHQnL0vd9neMrjZJ-Q,194623
+fbgemm_gpu/tbe/ssd/training.py,sha256=iepmavrK6cSXrqWg3TqVXL5gVmSftfHJycesp7I0Nw4,194911
 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-2025.9.29.dist-info/METADATA,sha256=EJ6RwlJa0HCpvnSKgLfI9lNQm95VHwHh9RosQp900Fg,2655
-fbgemm_gpu_genai_nightly-2025.9.29.dist-info/WHEEL,sha256=V2Q6mQKbouIadCxoRjt9FQ9oKfi45-uZUcoc77zzs0M,108
-fbgemm_gpu_genai_nightly-2025.9.29.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
-fbgemm_gpu_genai_nightly-2025.9.29.dist-info/RECORD,,
+fbgemm_gpu_genai_nightly-2025.10.2.dist-info/METADATA,sha256=KiCiRdg53J2HiyUZMdm_uIZHb-E8u0QQj9uRSc9oRIM,2655
+fbgemm_gpu_genai_nightly-2025.10.2.dist-info/WHEEL,sha256=V2Q6mQKbouIadCxoRjt9FQ9oKfi45-uZUcoc77zzs0M,108
+fbgemm_gpu_genai_nightly-2025.10.2.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
+fbgemm_gpu_genai_nightly-2025.10.2.dist-info/RECORD,,