fairseq2n is incompatible with torch 2.3.0

[0de554K]

Describe the bug: Attempted to bump torch version to 2.3.0 and got an error:

ERROR: fairseq2n 0.2.1 has requirement torch==2.2.2, but you'll have torch 2.3.0 which is incompatible

Describe how to reproduce: $python -m venv test $source test/bin/activate $pip install -r requirements_test_bug.txt

Content of requirements file: `torch==2.3.0; python_version >= '3.8' and sys_platform !="darwin"

#Seamless MT4 related packages fairseq2; sys_platform == "linux" and platform_machine != "aarch64" --extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/pt2.3.0/cu118 #fairseq2 dependency on libcudart.so.11.0 nvidia-cuda-runtime-cu11==11.7.99; sys_platform == "linux" and platform_machine != "aarch64" ` More other in case attempted to create a wheel it fails on dependency check

Describe the expected behavior: fairseq2n should be compatible with torch==2.3.0 as same as , according to docs, fairseq2 that seems to has dependency on the package.

Environment: $ python --version Python 3.8.10 $pip --version pip 20.0.2 $ pip freeze | grep torch torch==2.3.0 No CUDA d(i)GPU is used.

Additional Context: torch 2.3.0 is a preferable option for inferring of another model.

[gianpaj]

Same problem on NVIDIA Jetson Orin nano

 × No solution found when resolving dependencies:
  ╰─▶ Because fairseq2n==0.1.0 has no wheels with a matching platform tag (e.g., `manylinux_2_35_aarch64`) and
      fairseq2==0.1.0 depends on fairseq2n==0.1.0, we can conclude that fairseq2==0.1.0 cannot be used.
      And because only the following versions of fairseq2 are available:
          fairseq2==0.1.0
...
Because fairseq2n==0.4.6 has no wheels with a matching platform tag (e.g., `manylinux_2_35_aarch64`) and
      fairseq2==0.4.6 depends on fairseq2n==0.4.6, we can conclude that fairseq2==0.4.6 cannot be used.
      And because we know from (11) that fairseq2<0.4.6 cannot be used, we can conclude that all versions of fairseq2
      cannot be used.
      And because playdiffusion==0.1.0 depends on fairseq2, we can conclude that playdiffusion==0.1.0 cannot be used.
      And because only playdiffusion[demo]==0.1.0 is available and you require playdiffusion[demo], we can conclude
      that your requirements are unsatisfiable.

      hint: Wheels are available for `fairseq2n` (v0.1.0) on the following platform: `manylinux_2_17_x86_64`