rand BFloat16 sampling without conversion #73
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
fixes #71
Technically
randn
,randexp
should be written directly for BFloat16 too to avoid the rounding errors from conversion but they are 1) more complicted (see ziggurat algorithm for randn) 2) randn is continuous so rounding from conversion might be fine, randexp doesn't seem to produce very small numbers (<<1e-10) let alone 0 in practice for Float32/64, so rounding because of the conversion might be fine here tooEdit: I was wrong on
randexp
being complicated, from wikipedia