KaelanDt

Taking Quantum Geometric Tensor seriously

42

Hi, I noticed that when using an ARNN machine with complex parameters, computing the quantum geometric tensor using `netket.optimizer.qgt.QGTOnTheFly()` would yield a non-Hermitian matrix. Here is a minimal working example:...

Before submitting - [x] Was this discussed/approved via a Github issue? (no need for typos and docs improvements) - [x] Did you read the [contributor guideline](https://github.com/Lightning-AI/pytorch-lightning/blob/main/.github/CONTRIBUTING.md), Pull Request section? -...

## 📚 Documentation There are no docstrings for recipes and plugins, which are due to becoming the main entrypoint to thunder through thunder.compile (through https://github.com/Lightning-AI/lightning-thunder/pull/2096) For typos and doc fixes,...

## 🚀 Feature It would be better for the CI benchmarks: https://github.com/Lightning-AI/lightning-thunder/blob/main/.github/run-benchmark-as-lit-jobs.py are logged somewhere easily observable and accessible.

## 🚀 Feature As suggested by @Borda it would be nice to have automated litgpt benchmarks for thunder, in the same format as the HF model benchmarked through https://github.com/Lightning-AI/lightning-thunder/blob/main/.github/run-benchmark-as-lit-jobs.py. ###...

## What does this PR do? Fixes the benchmark job, see https://github.com/Lightning-AI/lightning-thunder/pull/2607/files. Required bumping `transformers` to run with `torch==2.8.0+cu12.8`, which fixes https://github.com/Lightning-AI/lightning-thunder/issues/2302 (although still not supporting vmap) ## PR review...