pytorch issue ，SequentialImpl can not push back DroupoutImpl

[mullerhai]

      private val sequential = new SequentialImpl()
      sequential.push_back(new DropoutImpl(dropout))

iceSupport] active device backend: cpu (requested=AUTO, cuda=false, mps=false, os=linux)
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x00007665b7153e37, pid=3256199, tid=3256200
#
# JRE version: Java(TM) SE Runtime Environment Oracle GraalVM 25+37.1 (25.0+37) (build 25+37-LTS-jvmci-b01)
# Java VM: Java HotSpot(TM) 64-Bit Server VM Oracle GraalVM 25+37.1 (25+37-LTS-jvmci-b01, mixed mode, sharing, tiered, jvmci, jvmci compiler, compressed oops, compressed class ptrs, g1 gc, linux-amd64)
# Problematic frame:
# C  [libjnitorch.so+0x1753e37]  torch::nn::AnyModuleHolder<torch::nn::DropoutImpl, at::Tensor>::forward(std::vector<torch::nn::AnyValue, std::allocator<torch::nn::AnyValue> >&&)+0x37
#
# Core dump will be written. Default location: Core dumps may be processed with "/usr/share/apport/apport -p%p -s%s -c%c -d%d -P%P -u%u -g%g -F%F -- %E" (or dumping to /home/muller/IdeaProjects/torch-rechub-scala/core.3256199)
#
# An error report file with more information is saved as:
# /home/muller/IdeaProjects/torch-rechub-scala/hs_err_pid3256199.log
[5.249s][warning][os] Loading hsdis library failed
#
# If you would like to submit a bug report, please visit:
#   https://bugreport.java.com/bugreport/crash.jsp
# The crash happened outside the Java Virtual Machine in native code.
# See problematic frame for where to report the bug.
#


# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x00007665b7153e37, pid=3256199, tid=3256200
#
# JRE version: Java(TM) SE Runtime Environment Oracle GraalVM 25+37.1 (25.0+37) (build 25+37-LTS-jvmci-b01)
# Java VM: Java HotSpot(TM) 64-Bit Server VM Oracle GraalVM 25+37.1 (25+37-LTS-jvmci-b01, mixed mode, sharing, tiered, jvmci, jvmci compiler, compressed oops, compressed class ptrs, g1 gc, linux-amd64)
# Problematic frame:
# C  [libjnitorch.so+0x1753e37]  torch::nn::AnyModuleHolder<torch::nn::DropoutImpl, at::Tensor>::forward(std::vector<torch::nn::AnyValue, std::allocator<torch::nn::AnyValue> >&&)+0x37
#
# Core dump will be written. Default location: Core dumps may be processed with "/usr/share/apport/apport -p%p -s%s -c%c -d%d -P%P -u%u -g%g -F%F -- %E" (or dumping to /home/muller/IdeaProjects/torch-rechub-scala/core.3256199)
#
# If you would like to submit a bug report, please visit:
#   https://bugreport.java.com/bugreport/crash.jsp
# The crash happened outside the Java Virtual Machine in native code.
# See problematic frame for where to report the bug.
#

---------------  S U M M A R Y ------------

Command Line: -XX:ThreadPriorityPolicy=1 -XX:+UnlockExperimentalVMOptions -XX:+EnableJVMCIProduct -XX:+EnableJVMCI -XX:-UnlockExperimentalVMOptions -javaagent:/snap/intellij-idea/86/lib/idea_rt.jar=33353 -Dfile.encoding=UTF-8 -Dsun.stdout.encoding=UTF-8 -Dsun.stderr.encoding=UTF-8 benchmarks.BenchmarkRunner

Host: 13th Gen Intel(R) Core(TM) i9-13900HX, 32 cores, 30G, Ubuntu 26.04 LTS
Time: Tue Jun 16 12:12:49 2026 CST elapsed time: 5.228522 seconds (0d 0h 0m 5s)

---------------  T H R E A D  ---------------

Current thread (0x000076987002f480):  JavaThread "main"             [_thread_in_native, id=3256200, stack(0x0000769874b00000,0x0000769874c00000) (1024K)]

Stack: [0x0000769874b00000,0x0000769874c00000],  sp=0x0000769874bfdc60,  free space=1015k
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
C  [libjnitorch.so+0x1753e37]  torch::nn::AnyModuleHolder<torch::nn::DropoutImpl, at::Tensor>::forward(std::vector<torch::nn::AnyValue, std::allocator<torch::nn::AnyValue> >&&)+0x37
Java frames: (J=compiled Java code, j=interpreted, Vv=VM code)
j  org.bytedeco.pytorch.SequentialImpl.forward(Lorg/bytedeco/pytorch/Tensor;)Lorg/bytedeco/pytorch/Tensor;+0
j  torchrec.basic.layers.MLP.forward(Lorg/bytedeco/pytorch/Tensor;)Lorg/bytedeco/pytorch/Tensor;+5
j  torchrec.models.ranking.LiquidNetWork.forward(Lscala/collection/immutable/Map;Lscala/collection/immutable/Map;)Lorg/bytedeco/pytorch/Tensor;+475
j  torchrec.trainers.CTRTrainer.fit$$anonfun$1(Ltorchrec/data/DataLoader;Lscala/Option;Ljava/lang/Object;I)V+1674
j  torchrec.trainers.CTRTrainer$$Lambda+0x0000000010387290.apply$mcVI$sp(I)V+17
J 2666 jvmci scala.runtime.java8.JFunction1$mcVI$sp.apply(Ljava/lang/Object;)Ljava/lang/Object; (14 bytes) @ 0x000076985e93c8a8 [0x000076985e93c3a0+0x0000000000000508]
J 2667 jvmci scala.collection.immutable.Range.foreach(Lscala/Function1;)V (43 bytes) @ 0x000076985e93fcf0 [0x000076985e93f180+0x0000000000000b70]
j  torchrec.trainers.CTRTrainer.fit(Ltorchrec/data/DataLoader;Lscala/Option;)V+42
j  benchmarks.BenchmarkRunner$.runLiquidNetWorkWithSequence(Lbenchmarks/BenchmarkConfig;Lscala/collection/immutable/List;Lscala/collection/immutable/List;)V+469
j  benchmarks.BenchmarkRunner$.runRankingBenchmark(Lbenchmarks/BenchmarkConfig;)Lbenchmarks/BenchmarkResult;+459
j  benchmarks.BenchmarkRunner$.runLiquidNetWorkBenchmark()Lbenchmarks/BenchmarkResult;+54
j  benchmarks.BenchmarkRunner$.main([Ljava/lang/String;)V+74
j  benchmarks.BenchmarkRunner.main([Ljava/lang/String;)V+4
v  ~StubRoutines::call_stub 0x000076985dfd2fa6

[mullerhai]

but if write code like that could work

      val dropoutLayer = new DropoutImpl(dropout)
      layersRef += dropoutLayer
      sequential.push_back(s"dropout_${layersRef.size}", dropoutLayer)

[mullerhai]

on the other test

  private val ffnHidden = 4 * dModel
  // 1. 显式声明并持有强引用，防止 GC 回收！
  private val ffnLinear1 = new LinearImpl(dModel, ffnHidden)
  private val ffnRelu = new ReLUImpl()
  private val ffnDrop1 = new DropoutImpl(dropout)
  private val ffnLinear2 = new LinearImpl(ffnHidden, dModel)
  private val ffnDrop2 = new DropoutImpl(dropout)

  private val ffn: SequentialImpl = {
    val seq = new SequentialImpl()
    // 2. 使用命名 push_back，传入已有强引用的层
    seq.push_back("ffn_lin1", ffnLinear1)
    seq.push_back("ffn_relu", ffnRelu)
    seq.push_back("ffn_drop1", ffnDrop1) //todo here will make crash
    seq.push_back("ffn_lin2", ffnLinear2)
   seq.push_back("ffn_drop2", ffnDrop2) //todo  here will make crash
    seq.to(targetDevice, false)
    seq
  }

#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x0000000000000000, pid=3551465, tid=3551466
#
# JRE version: Java(TM) SE Runtime Environment Oracle GraalVM 25+37.1 (25.0+37) (build 25+37-LTS-jvmci-b01)
# Java VM: Java HotSpot(TM) 64-Bit Server VM Oracle GraalVM 25+37.1 (25+37-LTS-jvmci-b01, mixed mode, sharing, tiered, jvmci, jvmci compiler, compressed oops, compressed class ptrs, g1 gc, linux-amd64)
# Problematic frame:
# C  [libjnitorch.so+0x1754092]  torch::nn::AnyModuleHolder<torch::nn::DropoutImpl, at::Tensor>::forward(std::vector<torch::nn::AnyValue, std::allocator<torch::nn::AnyValue> >&&)+0x292
#
# Core dump will be written. Default location: Core dumps may be processed with "/usr/share/apport/apport -p%p -s%s -c%c -d%d -P%P -u%u -g%g -F%F -- %E" (or dumping to /home/muller/IdeaProjects/torch-rechub-scala/core.3551465)
#
# If you would like to submit a bug report, please visit:
#   https://bugreport.java.com/bugreport/crash.jsp
# The crash happened outside the Java Virtual Machine in native code.
# See problematic frame for where to report the bug.
#

---------------  S U M M A R Y ------------

Command Line: -XX:ThreadPriorityPolicy=1 -XX:+UnlockExperimentalVMOptions -XX:+EnableJVMCIProduct -XX:+EnableJVMCI -XX:-UnlockExperimentalVMOptions -javaagent:/snap/intellij-idea/86/lib/idea_rt.jar=39137 -Dfile.encoding=UTF-8 -Dsun.stdout.encoding=UTF-8 -Dsun.stderr.encoding=UTF-8 benchmarks.BenchmarkRunner

Host: 13th Gen Intel(R) Core(TM) i9-13900HX, 32 cores, 30G, Ubuntu 26.04 LTS
Time: Tue Jun 16 16:43:21 2026 CST elapsed time: 9.160386 seconds (0d 0h 0m 9s)

---------------  T H R E A D  ---------------

Current thread (0x00007d29a402f480):  JavaThread "main"             [_thread_in_native, id=3551466, stack(0x00007d29a91da000,0x00007d29a92da000) (1024K)]

Stack: [0x00007d29a91da000,0x00007d29a92da000],  sp=0x00007d29a92d78b0,  free space=1014k
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
C  [libjnitorch.so+0x1754092]  torch::nn::AnyModuleHolder<torch::nn::DropoutImpl, at::Tensor>::forward(std::vector<torch::nn::AnyValue, std::allocator<torch::nn::AnyValue> >&&)+0x292
Java frames: (J=compiled Java code, j=interpreted, Vv=VM code)
j  org.bytedeco.pytorch.SequentialImpl.forward(Lorg/bytedeco/pytorch/Tensor;)Lorg/bytedeco/pytorch/Tensor;+0
j  torchrec.models.generative.HLLMTransformerBlock.forward(Lorg/bytedeco/pytorch/Tensor;Lscala/Option;)Lorg/bytedeco/pytorch/Tensor;+522
j  torchrec.models.generative.HLLM.forward$$anonfun$1(Lscala/runtime/ObjectRef;Lscala/Option;Ltorchrec/models/generative/HLLMTransformerBlock;)V+10
j  torchrec.models.generative.HLLM$$Lambda+0x000000007c3917c0.applyVoid(Ljava/lang/Object;)V+12
J 2754 c1 scala.runtime.function.JProcedure1.apply(Ljava/lang/Object;)Lscala/runtime/BoxedUnit; (11 bytes) @ 0x00007d298b2469cc [0x00007d298b246780+0x000000000000024c]
J 2753 c1 scala.runtime.function.JProcedure1.apply(Ljava/lang/Object;)Ljava/lang/Object; (8 bytes) @ 0x00007d298b24654c [0x00007d298b246300+0x000000000000024c]
J 3077 c1 scala.collection.immutable.List.foreach(Lscala/Function1;)V (32 bytes) @ 0x00007d298b2de36c [0x00007d298b2ddc80+0x00000000000006ec]
j  torchrec.models.generative.HLLM.forward(Lorg/bytedeco/pytorch/Tensor;Lscala/Option;)Lorg/bytedeco/pytorch/Tensor;+293
j  torchrec.trainers.CTRTrainer.fit$$anonfun$1(Ltorchrec/data/DataLoader;Lscala/Option;Ljava/lang/Object;I)V+2042
j  torchrec.trainers.CTRTrainer$$Lambda+0x000000007c37e170.apply$mcVI$sp(I)V+17
J 2752 jvmci scala.runtime.java8.JFunction1$mcVI$sp.apply(Ljava/lang/Object;)Ljava/lang/Object; (14 bytes) @ 0x00007d299294224c [0x00007d2992942100+0x000000000000014c]
j  scala.collection.immutable.Range.foreach(Lscala/Function1;)V+17
j  torchrec.trainers.CTRTrainer.fit(Ltorchrec/data/DataLoader;Lscala/Option;)V+42
j  benchmarks.BenchmarkRunner$.runGenerativeBenchmark(Lbenchmarks/BenchmarkConfig;)Lbenchmarks/BenchmarkResult;+1324
j  benchmarks.BenchmarkRunner$.runHLLMBenchmark()Lbenchmarks/BenchmarkResult;+54
j  benchmarks.BenchmarkRunner$.main([Ljava/lang/String;)V+74
j  benchmarks.BenchmarkRunner.main([Ljava/lang/String;)V+4
v  ~StubRoutines::call_stub 0x00007d2991fd2fa6

[mullerhai]

I think SequentialImpl maybe have more bug ,we need to solve it

[saudet]

It just sounds like the objects get deallocated too early.

Please try to set the "org.bytedeco.javacpp.nopointergc" system property to "true".