Introducing quantized_linear #118
Conversation
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
mschoenb97
changed the title
|
Since this deprecates |
|
|
||
| def test_GetScale_PerChannelScale(): | ||
| # Rank1 tensors | ||
| x_r1 = tf.ones([4]) |
There was a problem hiding this comment.
Since quantized_linear is trying to incorporate binary quantizer as well, I suggest add more scale test here for per channel scales.
- instead of having input all as 1, consider use wider range of input values so we can test better how scale is calculated;
- Add test on quantizer scale during training and eval; to make sure scales don't change between two stages.
There was a problem hiding this comment.
This is not a test for quantized_linear. I only made a change here because I changed the name of the _get_scale function.
However, these are still good points, and I will add tests with specific scale values to the rest of the quantized_linear tests in theqactivations_test.py file
There was a problem hiding this comment.
Added both (1) and (2)!
| scale_r4_eps_2_ua_3 = quantizers._get_least_squares_scale( | ||
| "auto", x_r4, q_r4, elements_per_scale=2, scale_axis=3) | ||
|
|
||
| assert_equal(tf.shape(scale_r4_eps_none_ua_none).numpy(), [1, 1, 1, 16]) |
There was a problem hiding this comment.
consider add assertion on some specific scale values as well. The same goes for other related tests.
There was a problem hiding this comment.
See discussion above
4 participants
New quantizer to replace quantized_bits. Summary of improvements from old version: