Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions torchao/float8/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,8 @@ throughput speedups of up to 1.5x on 128 GPU LLaMa 3 70B pretraining jobs.

:warning: <em>The codebase is stable, but backwards compatibility is not yet guaranteed.</em>

:warning: <em>These APIs are training-only and float8-only, and we plan to [unify them with the rest of torchao](https://github.com/pytorch/ao/issues/894) in the future.</em>

# Single GPU User API

We provide three per-tensor scaling strategies: dynamic, delayed and static. See https://arxiv.org/pdf/2209.05433.pdf, Section 4.3 for more details. These strategies are configurable separately for activations (`input`), weights (`weight`) and gradients (`grad_output`).
Expand Down
Loading