Benchmark: Model benchmark - deterministic training support#731

Open

Aishwarya-Tonpe wants to merge 97 commits intomainfrom

aishwaryatonpe/deterministic-training

Commits on Aug 17, 2025

Add deterministic training functionality to PyTorch LLaMA benchmark
Aishwarya-Tonpe
committed
llama: add periodic checksum logging (deterministic-only, log-only); pass check_frequency to _is_finished in train/infer; add test capturing checksum log; stabilize fp32 loss path and small-dims de…
Aishwarya-Tonpe
committed
deterministic training: enable seeding + deterministic algorithms across BERT/GPT2/CNN/LSTM/Mixtral; per-step fp32 loss logging; checksum logs; tests updated to strict/soft determinism pattern; add…
Aishwarya-Tonpe
committed
tests(pytorch): add strict determinism skip guards and detailed docstrings; fix GPT-2 params; soft vs strict checks stabilized
Aishwarya-Tonpe
committed
Refactor LLaMA model tests: align strict, soft determinism, and checksum tests with BERT pattern, improve docstrings and skip logic.
Aishwarya-Tonpe
committed
examples: add deterministic and strict_determinism flags and docs to BERT, GPT-2, LSTM, CNN, LLaMA examples
Aishwarya-Tonpe
committed
Deterministic fingerprints: replace checksum with Loss+ActMean across models; update tests
Aishwarya-Tonpe
committed
Deterministic training + reproducible logging: align GPT-2/LLaMA/LSTM/CNN/BERT/Mixtral with periodic fingerprints, per-step loss capture, TF32 off, SDPA math kernel; add model_log_utils; update exa…
Aishwarya-Tonpe
committed

Commits on Aug 18, 2025

Adding flag: Checck-frequency
Aishwarya-Tonpe
committed

Commits on Aug 19, 2025

Add Check frequency flag to tests
Aishwarya-Tonpe
committed

Commits on Aug 20, 2025

Commits on Aug 21, 2025

Commits on Aug 26, 2025

Add Docstrings, align imports, add assertions messages
Aishwarya-Tonpe
committed

Commits on Aug 27, 2025

Lint Checks
Aishwarya-Tonpe
committed

Commits on Sep 3, 2025

Removing unnecessary code
Aishwarya-Tonpe
committed

Commits on Sep 15, 2025

Commits on Sep 22, 2025

Merge branch 'main' into aishwaryatonpe/deterministic-training
polarG
authored

Commits on Oct 1, 2025

Commits on Oct 2, 2025

Commits on Oct 8, 2025

Commits on Oct 9, 2025

Commits on Oct 23, 2025

Merge branch 'main' into aishwaryatonpe/deterministic-training
Aishwarya-Tonpe
authored

Commits on Nov 10, 2025

Merge branch 'main' into aishwaryatonpe/deterministic-training
Aishwarya-Tonpe
authored

Commits on Dec 8, 2025

Merge branch 'main' of https://github.com/microsoft/superbenchmark into aishwaryatonpe/deterministic-training

root
authored and
root
committed
Removing check_frequency parameter from is_finished method in train and inference steps

root
authored and
root
committed
Comments resolve : Removing check_frequency assignment to the variable

root
authored and
root
committed
Update superbench/benchmarks/model_benchmarks/pytorch_base.py

Aishwarya-Tonpe
and
Copilot
authored
Update tests/benchmarks/model_benchmarks/test_pytorch_determinism_all.py

Aishwarya-Tonpe
and
Copilot
authored
Update superbench/benchmarks/model_benchmarks/pytorch_base.py

Aishwarya-Tonpe
and
Copilot
authored

Commits on Dec 15, 2025

Commits on Dec 16, 2025

Commits on Dec 19, 2025

Minor change
Aishwarya-Tonpe
committed

Commits on Jan 6, 2026

Merge branch 'main' into aishwaryatonpe/deterministic-training
Aishwarya-Tonpe
authored