Benjamin Bastian

PyTorch/XLA should not crash runtime during int64 dot product on TPU

6

## Issue description Currently, PyTorch/XLA will crash the entire runtime (ret check failure?) if you do an int64 dot product on a TPU. Seems like poor UX to me. It'd...

### Description JAX doesn't nicely share TPUs with other frameworks (e.g. PyTorch/XLA, TF, etc). This is fine, but the error reported by JAX is misleading. It'd be preferable to actively...

BF16 matmul slower than F32 matmul on T4 GPU

3

T4 GPU doesn't support BF16 matmul. Because of this, XLA switches BF16 matmul to F32 matmul on T4 (IIUC). This is obviously much slower, but it turns out it's actually...

BF16 matmul slower than F32 matmul on T4 GPU

3

### Description BF16 matmul appears to be slower than F32 matmul on T4. From my test, BF16 appears to be half the speed. I believe this is a bug and...

Some incantations of find-and-replace cause monaco-vim to hang indefinitely. You can reproduce this with `:%s/$/,`. I was able to verify [here](https://editor.bitwiser.in/) that this happens in the most recent version of...

`pip install torchtext` is broken after torch 2.4.0

5

## 🐛 Bug `torchtext` is in maintenance mode, but there's a problem with the current dependencies which I think may warrant an update and minor version bump. This problem causes...

- Add log-space gaussian probability function using logsumexp - Clamp sigma to prevent division by zero (MIN_SIGMA=1e-6) - Fix device/dtype handling in sample() for GPU compatibility - Remove unused imports...