This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* AMP improvements + enable bf16 input for quantize_v2 * Fix sanity * Improve tests, AMP conversion interface, fix forwad hooks * Fix tests * Fix imports in tests * Use different lp16_fp32 op in test * Add amp.disable_amp() context, fix tests * Add tests, generalize optimization disabling * Fix sanity * Review fixes * Use is_integral<>::value * Review fixes Change flag type to unsigned int Add a warning for an incorrect flag attribute value * Extend bf16 support * Combine enable_float_output and amp_out_dtype parameters * Add bf16 support to _dnnl_batch_dot * Fix sanity * Add bf16 support to all dnnl ops, add tests * Add license * Fix conv activation fuse, disable masked_softmax bf16 support * Fix sanity, add softmax test cases * Compare bf16 outputs with fp32 reference Co-authored-by: Bartlomiej Gawrych <[email protected]>
- Loading branch information