apex @ nemo v2.2.1 version: build fails: ATen/Dispatch.h:199:48: error: cannot convert ‘const at::DeprecatedTypeProperties’ to ‘c10::ScalarType’

[eugeneswalker]

Describe the Bug

I'm trying to build the version of Apex used in NeMo v2.2.1 -- 810ffae -- as listed here:

• https://github.com/NVIDIA/NeMo/tree/v2.2.1?tab=readme-ov-file#install-llms-and-mms-dependencies

I'm configuring the build with the options listed here:

• https://github.com/NVIDIA/NeMo/tree/v2.2.1?tab=readme-ov-file#apex

--build-option=--cpp_ext --cuda_ext --fast_layer_norm --distributed_adam --deprecated_fused_adam --group_norm

The build is failing in a number of different environments with this error:

...
  /usr/local/lib/python3.12/dist-packages/torch/include/ATen/Dispatch.h:199:48: error: cannot convert ‘const at::DeprecatedTypeProperties’ to ‘c10::ScalarType’
    199 |     at::ScalarType _st = ::detail::scalar_type(the_type);                   \
        |                                                ^~~~~~~~
        |                                                |
        |                                                const at::DeprecatedTypeProperties

Minimal Steps/Code to Reproduce the Bug

The error can be reliably reproduced inside nvcr.io/nvidia/pytorch:25.02-py3:

• Build log: output.txt

$> docker run -it nvcr.io/nvidia/pytorch:25.02-py3

root@3adeea6b839b:/# mkdir -p /repos
root@3adeea6b839b:/# cd /repos
root@3adeea6b839b:/repos# pip install packaging build
root@3adeea6b839b:/repos# git clone https://github.com/NVIDIA/apex
root@3adeea6b839b:/repos# cd apex
root@3adeea6b839b:/repos/apex# export apex_commit=810ffae374a2b9cb4b5c5e28eaeca7d7998fca0c
root@3adeea6b839b:/repos/apex# pip wheel -v --no-build-isolation \
 --disable-pip-version-check \
 --no-cache-dir \
 --config-settings "--build-option=--cpp_ext --cuda_ext --fast_layer_norm --distributed_adam --deprecated_fused_adam --group_norm" . \
  2>&1 | tee output
...
  /usr/local/lib/python3.12/dist-packages/torch/include/ATen/Dispatch.h:199:48: error: cannot convert ‘const at::DeprecatedTypeProperties’ to ‘c10::ScalarType’
    199 |     at::ScalarType _st = ::detail::scalar_type(the_type);                   \
        |                                                ^~~~~~~~
        |                                                |
        |                                                const at::DeprecatedTypeProperties
  /usr/local/lib/python3.12/dist-packages/torch/include/ATen/Dispatch.h:227:3: note: in expansion of macro ‘AT_DISPATCH_SWITCH’
    227 |   AT_DISPATCH_SWITCH(                                        \
        |   ^~~~~~~~~~~~~~~~~~
  /repos/apex/csrc/mlp.cpp:69:3: note: in expansion of macro ‘AT_DISPATCH_FLOATING_TYPES_AND_HALF’
     69 |   AT_DISPATCH_FLOATING_TYPES_AND_HALF(inputs[0].type(), "mlp_forward", [&] {
...

Expected Behavior
I expect the build to succeed and a wheel to be produced.

Environment

PyTorch version: 2.7.0a0+ecf3bae40a.nv25.02
Is debug build: False
CUDA used to build PyTorch: 12.8
ROCM used to build PyTorch: N/A

OS: Ubuntu 24.04.1 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: Could not collect
CMake version: version 3.31.4
Libc version: glibc-2.39

Python version: 3.12.3 (main, Jan 17 2025, 18:03:48) [GCC 13.3.0] (64-bit runtime)
Python platform: Linux-4.18.0-553.16.1.el8_10.x86_64-x86_64-with-glibc2.39
Is CUDA available: True
CUDA runtime version: 12.8.61
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA H100 PCIe
Nvidia driver version: 570.124.06
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.7.1
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.7.1
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                         x86_64
CPU op-mode(s):                       32-bit, 64-bit
Address sizes:                        48 bits physical, 48 bits virtual
Byte Order:                           Little Endian
CPU(s):                               96
On-line CPU(s) list:                  0-95
Vendor ID:                            AuthenticAMD
Model name:                           AMD EPYC 7413 24-Core Processor
CPU family:                           25
Model:                                1
Thread(s) per core:                   2
Core(s) per socket:                   24
Socket(s):                            2
Stepping:                             1
Frequency boost:                      enabled
CPU(s) scaling MHz:                   66%
CPU max MHz:                          3630.8101
CPU min MHz:                          1500.0000
BogoMIPS:                             5300.06
Flags:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr wbnoinvd amd_ppin brs arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm
Virtualization:                       AMD-V
L1d cache:                            1.5 MiB (48 instances)
L1i cache:                            1.5 MiB (48 instances)
L2 cache:                             24 MiB (48 instances)
L3 cache:                             256 MiB (8 instances)
NUMA node(s):                         2
NUMA node0 CPU(s):                    0-23,48-71
NUMA node1 CPU(s):                    24-47,72-95
Vulnerability Gather data sampling:   Not affected
Vulnerability Itlb multihit:          Not affected
Vulnerability L1tf:                   Not affected
Vulnerability Mds:                    Not affected
Vulnerability Meltdown:               Not affected
Vulnerability Mmio stale data:        Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Not affected
Vulnerability Spec rstack overflow:   Mitigation; Safe RET
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

Versions of relevant libraries:
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] nvidia-cudnn-frontend==1.10.0
[pip3] nvtx==0.2.5
[pip3] onnx==1.17.0
[pip3] optree==0.14.0
[pip3] pynvjitlink==0.3.0
[pip3] pytorch-triton==3.2.0+git0d4682f0b.nvinternal
[pip3] torch==2.7.0a0+ecf3bae40a.nv25.2
[pip3] torch_geometric==2.5.3
[pip3] torch_tensorrt==2.6.0a0
[pip3] torchprofile==0.0.4
[pip3] torchvision==0.22.0a0
[conda] Could not collect

[eqy]

The 25.02 PyTorch container has its own version of Apex that is built and validated against the PyTorch install present---I'm not sure there's an expectation that NeMo's pinned version of Apex would be compatible with 25.02's PyTorch version.

Have you tried using the NeMo NGC container directly? https://catalog.ngc.nvidia.com/orgs/nvidia/containers/nemo/tags

[eugeneswalker]

Thank you for your help!

The 25.02 PyTorch container has its own version of Apex that is built and validated against the PyTorch install present---I'm not sure there's an expectation that NeMo's pinned version of Apex would be compatible with 25.02's PyTorch version.

Do you know if the version/fork/commit of apex they are using is documented anywhere? I just need a version of apex that can be built+configured to work properly w nemo 2.2.1 and recent pytorch.

Have you tried using the NeMo NGC container directly? https://catalog.ngc.nvidia.com/orgs/nvidia/containers/nemo/tags

Yes, the container works great. But, I am working on a project where we are doing our own containerized build of Nemo framework. So, we are trying to follow documented install instructions and running into this.

[eqy]

Thank you for your help!

The 25.02 PyTorch container has its own version of Apex that is built and validated against the PyTorch install present---I'm not sure there's an expectation that NeMo's pinned version of Apex would be compatible with 25.02's PyTorch version.

Do you know if the version/fork/commit of apex they are using is documented anywhere? I just need a version of apex that can be built+configured to work properly w nemo 2.2.1 and recent pytorch.

Have you tried using the NeMo NGC container directly? https://catalog.ngc.nvidia.com/orgs/nvidia/containers/nemo/tags

Yes, the container works great. But, I am working on a project where we are doing our own containerized build of Nemo framework. So, we are trying to follow documented install instructions and running into this.

For the PyTorch NGC container (which I think is the relevant one here as that's where the build error seems to be coming from) the source tree for the version of apex used should just be in the container in /opt/pytorch/apex.

[eugeneswalker]

For the PyTorch NGC container (which I think is the relevant one here as that's where the build error seems to be coming from) the source tree for the version of apex used should just be in the container in /opt/pytorch/apex.

Ahh! Excellent. It built fine when I used the source tree bundled with the image. Should we close this issue then?

Do you know what checkout they use for /opt/pytorch/apex- the git info looks to have been deleted in the image.