-
### Feature details
In #4585, the decomposition of `TmpPauliRot`, a helper object of `SpecialUnitary`, was changed in order to allow the new `DefaultQubit` device to differentiate `SpecialUnitary`.
…
-
**Is your feature request related to a problem? Please describe.**
I have a special use case for my buffer pool,
```rust
struct BufferPool {
keep_capacity: usize,
buffers: UnsafeCell,
}…
-
In a common use case:
- train the model on a small sample of data,
- save the weight table,
- and then reuse the learned weights for inference on a larger dataset.
When doing the inference, many f…
-
---
Author Name: **Peter Nordin** (@peterNordin)
Original Redmine Issue: 481, https://flumes.iei.liu.se/redmine/issues/481
Original Date: 2012/03/20
---
Abort or warn if divide by zero
Right n…
-
The latter can run in parallel with dfun eval, something like
``` c
#pragma omp parallel sections
{
#pragma omp section
{
(*hist->get_next_non_zero)(hist, c);
}
#pragma omp s…
-
right now `infer()` only works with `Type`s not `TypeRef`s.
The only challenge here is in the App case when you construct `t1 -> t2` then unify that with something else. The issue is t1 and t2 migh…
-
We need to setup a system that allows painless split-testing.
We must not be inhibited to split-test features, no matter how trivial. Therefore the system must come at next-to-zero added overhead to …
-
Create a new `distributed_train` function in `torchplate.experiment.Experiment` which interfaces with [Hugging Face Accelerate](https://huggingface.co/docs/accelerate/quicktour) for zero-overhead dist…
-
Realm provides [a profiling response for measuring the upper and lower bound on when kernels launched by a GPU task are executed](https://gitlab.com/StanfordLegion/legion/-/blob/master/runtime/realm/p…
-
Dispatch time can be a limiting factor for perf, especially for decode, where op device latency are low. We show a traced Mixtral decode perf below:
If dispatch went to 0, perf boost would be: 1.…