fbgemm-gpu-genai-nightly 2025.10.16__cp310-cp310-manylinux_2_28_x86_64.whl → 2025.10.18__cp310-cp310-manylinux_2_28_x86_64.whl

fbgemm_gpu/docs/target.genai.json.py

@@ -1,6 +1,6 @@
 
 {
-    "version": "2025.10.16",
+    "version": "2025.10.18",
     "target": "genai",
     "variant": "cuda"
 }

fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py

@@ -175,7 +175,7 @@ class QuantizeOpBase(metaclass=abc.ABCMeta):
             if use_cuda_graph:
                 with torch.cuda.stream(torch.cuda.Stream()):
                     t = triton.testing.do_bench_cudagraph(
-                        lambda: self.quantize_and_compute(*args, **kwargs)
+                        lambda: self.quantize_and_compute(*args, **kwargs), rep=200
                     )
             else:
                 t = triton.testing.do_bench(
@@ -188,7 +188,7 @@ class QuantizeOpBase(metaclass=abc.ABCMeta):
                 if use_cuda_graph:
                     with torch.cuda.stream(torch.cuda.Stream()):
                         t = triton.testing.do_bench_cudagraph(
-                            lambda: self.compute(*args, **kwargs)
+                            lambda: self.compute(*args, **kwargs), rep=200
                         )
                 else:
                     t = triton.testing.do_bench(lambda: self.compute(*args, **kwargs))

fbgemm_gpu/split_table_batched_embeddings_ops_training.py

@@ -1714,10 +1714,16 @@ class SplitTableBatchedEmbeddingBagsCodegen(nn.Module):
             tbe_id=self.uuid,
         )
 
+    def _get_tensor_memory(self, tensor_name: str) -> int:
+        """Get memory usage of a tensor in bytes."""
+        if not hasattr(self, tensor_name):
+            self.log(f"Tensor '{tensor_name}' not found, using 0 bytes")
+            return 0
+        tensor = getattr(self, tensor_name)
+        return tensor.numel() * tensor.element_size()
+
     @torch.jit.ignore
-    def _report_tbe_mem_usage(
-        self,
-    ) -> None:
+    def _report_tbe_mem_usage(self) -> None:
         if self.stats_reporter is None:
             return
 
@@ -1726,18 +1732,17 @@ class SplitTableBatchedEmbeddingBagsCodegen(nn.Module):
             return
 
         total_mem_usage = sum(
-            param.numel() * param.element_size() for param in self.parameters()
-        ) + sum(buffer.numel() * buffer.element_size() for buffer in self.buffers())
+            p.numel() * p.element_size() for p in self.parameters()
+        ) + sum(b.numel() * b.element_size() for b in self.buffers())
+
         if self.use_cpu:
             total_hbm_usage = 0
             total_uvm_usage = total_mem_usage
         else:
-            # hbm usage is total usage minus uvm usage
             total_uvm_usage = sum(
-                getattr(self, tensor_name).numel()
-                * getattr(self, tensor_name).element_size()
-                for tensor_name in self._uvm_tensors_log
-                if hasattr(self, tensor_name)
+                self._get_tensor_memory(name)
+                for name in self._uvm_tensors_log
+                if hasattr(self, name)
             )
             total_hbm_usage = total_mem_usage - total_uvm_usage
 

{fbgemm_gpu_genai_nightly-2025.10.16.dist-info → fbgemm_gpu_genai_nightly-2025.10.18.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fbgemm_gpu_genai_nightly
-Version: 2025.10.16
+Version: 2025.10.18
 Home-page: https://github.com/pytorch/fbgemm
 Author: FBGEMM Team
 Author-email: packages@pytorch.org

{fbgemm_gpu_genai_nightly-2025.10.16.dist-info → fbgemm_gpu_genai_nightly-2025.10.18.dist-info}/RECORD

@@ -1,8 +1,8 @@
 fbgemm_gpu/__init__.py,sha256=A3DuseilQ-sEtBpeZsG0LOqN5Cl3e5DHI_YgCZEMhnE,6283
-fbgemm_gpu/asmjit.so,sha256=5N9owjQJb6tzArT7Lzjl_gIEjn5eDTvOvjxwLEVTaV4,484232
+fbgemm_gpu/asmjit.so,sha256=s2majpI_vsFgc4Vh8szCK4xFYXyT5A0K6HwepGBwrqQ,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
@@ -17,7 +17,7 @@ fbgemm_gpu/split_embedding_utils.py,sha256=Gb40ZKeATxIKEKI3aVQMgDDBanNpKMc53Z43m
 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_inference.py,sha256=dGC85xjQiRUrequBibSf9oMAVHT5Q49zsVo2zW4n_88,81679
-fbgemm_gpu/split_table_batched_embeddings_ops_training.py,sha256=lF9eP6GDTyqbEJgl-SO6gNYUk2dv2YE2bMEtzGkY21c,173757
+fbgemm_gpu/split_table_batched_embeddings_ops_training.py,sha256=nbrdPt2WYeVB1BDyToa4vfl_XiOza5dEGjSB8jCV_mY,173930
 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=YLp8TDJT2TCspqkyamSM_Q4YhyBNhIAUhSrX5oQIkQ8,79
+fbgemm_gpu/docs/target.genai.json.py,sha256=65V25E8odqOdgvGWjS63CwjMVYLsXRoJAK8-iaPYW7g,79
 fbgemm_gpu/experimental/example/__init__.py,sha256=OvJHZgWnycL1gWKyCXFJCTKuys3KAqx4iadjx3R-tBQ,723
-fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=VqjkB_CVMV_gQLGn9Upu-2SwrpeReWK43f0pqsYnWhs,232488
+fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so,sha256=PGtZj3tM9mq65PGD08gEiTlj5PsvGaqJ_VkCvveHIIk,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=KWiKRpQI9j6UCasInw0lgqYAr1bKLY51pzysGdd_9Go,72663336
+fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so,sha256=L6aV0GcM8BKbNcDdZJAIgm4BLBspjo4MQ2em452K_ag,73184824
 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
@@ -52,7 +52,7 @@ fbgemm_gpu/experimental/gen_ai/bench/__init__.py,sha256=XpAK_eyqDSKeFC5J9KpnKtbZ
 fbgemm_gpu/experimental/gen_ai/bench/comm_bench.py,sha256=ApEyJOf_rdIo8V_EgvhZXBGNov8ITC_dnB95v8szulI,8515
 fbgemm_gpu/experimental/gen_ai/bench/gather_scatter_bench.py,sha256=K9Nib6D7xJbw1QwEVuCJrVyI1qs988moo3cieVKYuFY,12057
 fbgemm_gpu/experimental/gen_ai/bench/quantize_bench.py,sha256=BWl6t-4acbuRSEX2aVNDlFrSWZkqMWK2sI3VONaMd3Q,24047
-fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py,sha256=H6AchejyZs76_snM_ae5vV0cPr_Q0h35OQ8qED0r1N4,104915
+fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py,sha256=cDZS2rCb1W2IEQYxsnGjauhlUhg2PFZ-9LqJ_SEdbiQ,104933
 fbgemm_gpu/experimental/gen_ai/moe/README.md,sha256=z9ybHmv4KFJ1drj5OByuFaOY0tRQwwiIW3Q22TB_2-k,904
 fbgemm_gpu/experimental/gen_ai/moe/__init__.py,sha256=lwSvff07yEav024B1XyfgW8r8hwNe--aEDywcO7rnbM,1905
 fbgemm_gpu/experimental/gen_ai/moe/activation.py,sha256=NiXhWyCNagI3P9N3N89iSX7xKuShdkq9DxEUAzoV6y0,7892
@@ -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-2025.10.16.dist-info/METADATA,sha256=94b2O3y17SKbTnbPh3zt-kbAQX8GRQpAY3z6-ZJ48uo,2656
-fbgemm_gpu_genai_nightly-2025.10.16.dist-info/WHEEL,sha256=k9CVMKlTmOLLXq_OyiiJFbPd6UKfogV4yIUezgPmplE,108
-fbgemm_gpu_genai_nightly-2025.10.16.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
-fbgemm_gpu_genai_nightly-2025.10.16.dist-info/RECORD,,
+fbgemm_gpu_genai_nightly-2025.10.18.dist-info/METADATA,sha256=sUOrHh3PBHZdQr2NaPSJGmkTiw9K3N9oYAO6xvNYqyg,2656
+fbgemm_gpu_genai_nightly-2025.10.18.dist-info/WHEEL,sha256=k9CVMKlTmOLLXq_OyiiJFbPd6UKfogV4yIUezgPmplE,108
+fbgemm_gpu_genai_nightly-2025.10.18.dist-info/top_level.txt,sha256=_2s1Aa08r_eDn0JP4FjOhzK09Q8bVlEI7q8pMep51UY,25
+fbgemm_gpu_genai_nightly-2025.10.18.dist-info/RECORD,,