[HowieYin]

Summary

Half sampling accumulates softmax and CDF in FP16. With 32768 zero logits, probabilities sum to 16 instead of 1; seed 42 returns token 694 instead of the FP32 result, 11111.

Use the existing FP32 buffer, as in the BF16 path (#22094), for temperature scaling, softmax and sampling. Keep greedy sampling on the original logits.

Test plan

Locally compiled test_sampler.cpp with libtorch and GoogleTest on macOS arm64: ./sampler_test — 11 passed. The three new large-vocabulary tests fail before the fix and pass after it, covering multinomial, top-p and top-k over repeated calls.

Portable C++17 build with ASan/UBSan: ./portable_sampler_test — 3 passed. clang-format 18.1.3 and git diff --check passed.

AI assistance: TRAE.

cc @larryliu0820 @mergennachin @cccclai @helunwencser @jackzhxng @digantdesai

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24021

• 📄 Preview Python docs built from this PR

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

⚠️ 24 Awaiting Approval

As of commit e786a10 with merge base 03ddd8b ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

• Build Aarch64 Linux CUDA Wheels (gh)
• Build Aarch64 Linux Wheels (gh)
• Build documentation (gh)
• Build Linux CUDA Wheels (gh)
• Build Linux Wheels (gh)
• Build macOS Wheels (gh)
• Build Presets (gh)
• Build Windows CUDA Wheels (gh)
• Build Windows Wheels (gh)
• Cadence Build & Test (gh)
• Lint (gh)
• MLX (gh)
• periodic (gh)
• pull (gh)
• Test ARM Backend (gh)
• Test CoreML Backend (gh)
• Test Cortex-M Backend (gh)
• Test NXP Backend (gh)
• Test QNN Backend (gh)
• Test QNN Windows MSVC build (gh)
• Test Vulkan Backend (gh)
• Test Vulkan Backend (specialized runners) (gh)
• Test WebGPU Backend (gh)
• Test XNNPACK Backend (gh)

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

[meta-cla]

Hi @HowieYin!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[linux-foundation-easycla]

• ❌ - login: @HowieYin / name: HowieYin. The commit (e786a10) is not authorized under a signed CLA. Please click here to be authorized. For further assistance with EasyCLA, please visit our EasyCLA portal and chat with our support bot.

[HowieYin]

@pytorchbot label "release notes: llm" "module: llm" "bug"