Ivan Yashchuk

> `var` is intentionally left as a primitive so that backends are free to implement one pass welford approach. I thought backends in general should be able to opt-in to...

No, it doesn't need it.

Can we make `var_mean` be prim for the same reason `var` is prim? Would it be accepted?

SVD: https://j-towns.github.io/papers/svd-derivative.pdf

The problem is in `torch.cholesky_solve`. Modifying GPyTorch source at `models/exact_prediction_strategy.py` [here](https://github.com/cornellius-gp/gpytorch/blob/fc2053b0fc00517880fbc11adc7f5802242eec6a/gpytorch/models/exact_prediction_strategies.py#L314) "fixes" the problem. ``` - covar_correction_rhs = train_train_covar.inv_matmul(train_test_covar) + covar_correction_rhs = train_train_covar.to('cpu').inv_matmul(train_test_covar.to('cpu')).to('cuda') ``` Here's a minimal reproducer of the...

> Is "tangent linear method" a termed used in the certain fields? It is a common term, even used in the current draft https://github.com/bob-carpenter/ad-handbook/blob/09c85566defab0c78fb60a7913425166be6b2e0f/index.Rmd#L103 > the case of PDEs is...

I think I've addressed all the needed changes. Rendered latex derivation: https://github.com/bob-carpenter/ad-handbook/pull/8#discussion_r409607668

Rebuild for Linux will be triggered automatically after the merge, right? https://github.com/pytorch/builder/blob/main/.github/workflows/build-magma-linux.yml

@atalman, could you please approve running workflows?

Hello, @aravindshaj! Unfortunately, `jax.jit` is not supported yet. Previously, to make it work [XLA CustomCall](https://www.tensorflow.org/xla/custom_call) needed to be used. Now it seems like it's not difficult to add support of...