Results 191 issues of Lukasz Stafiniak

In the current giant refactor, I'm removing dynamic indexing. Also perhaps simplifying the "local / global on device / on host" behavior.

explore

Exact interpretation filling in the "metadata" is very slow.

enhancement

It would probably mean emitting different instructions in `Ndarray`. Propagating the setting in `Formula` might be tricky since it needs to be partially top-down.

This improves consistency and avoids bugs when printing `%d` vs. `%u`. Rewriting `Exec_as_cuda` currently.

Needs to be fully integrated with shape inference. The core functionality would be around an extension of the `einsum` notation, this allows to align and permute axes arbitrarily (on top...

enhancement

Needed for completeness but unlikely to manifest.

Globals are convenient but bad, make them less bad.

enhancement

It is out-of-scope for this issue to implement support for open-weights models such as GPT-2 or Gemma: I'm leaving that for version 0.7. We should avoid doing work on a...

explore

The dataset consisting of (positive, negative) output pairs, and a Human Feedback UI to grow it via inference on the model while the model is training.

explore