add gptq benchmark, and speed up by ~3x with compile

[vkuzo]

Summary:

1. add gptq algo benchmark
2. capture a profiling run - two main issues, (a) the to_nvfp4 kernels and
   (b) the hessian + weight updates
3. speed up (a) significantly by using torch.compile, we need to
   ensure to enable TORCHINDUCTOR_EMULATE_DIVISION_ROUNDING to keep
   parity with eager mode. Note there are significant accuracy
   regressions without this set (small differences accumulate with each
   incremental hessian + weight update), so we force it on by default.

e2e gptq run on LLaMa 3.2 1B speeds up by ~2.5x (172 seconds -> 69 seconds) with no change in accuracy:

time torchao/prototype/gptq/gptq_nvfp4_llama3_2_1b_nonsequential_wikitext.sh 2>&1 | tee ~/tmp/20260421_gptq_run_compile.txt
// logs: https://gist.github.com/vkuzo/4ebd0bfa71185b711b19c712216eae29

Test Plan:

[vkuzo]

Stack from ghstack (oldest at bottom):

• hook up real nvfp4 grouped_gemm #4316
• make NVFP4Tensor handle per-expert outer scale #4315
• emulated nvfp4 support torch._grouped_mm for inference #4314
• gptq example: remove transformers version check #4313
• -> add gptq benchmark, and speed up by ~3x with compile #4310

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4310

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

❌ 1 New Failure

As of commit d302888 with merge base 1b04df3 ():

NEW FAILURE - The following job has failed:

• Run 1xH100 Tests / test (H100, linux.aws.h100, --pre torch torchvision torchaudio mslk --index-url https://download.... / linux-job (gh)
  RuntimeError: Command docker exec -t e92b6a762ea62a740525282623b6cd68d12412f1a28ec5f150f5fc07b7eee2b1 /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.