Add option to disable fake quant for 8da4w QAT #198
Conversation
facebook-github-bot
added
the
CLA Signed
andrewor14
force-pushed
the
8da4w_qat_enable
branch
from
4f3b17e
to
25c36be
Compare
|
Actually I had to remove |
andrewor14
force-pushed
the
8da4w_qat_enable
branch
from
25c36be
to
fbc5742
Compare
|
@andrewor14 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
Summary: This feature helps with model convergence during QAT. The user can disable observation/fake quant for the first N steps and renable them later, allowing the activation and weight values to stabilize before applying quantization. Test Plan: python test/quantization/test_qat.py -k test_qat_8da4w_quantizer_disable_fake_quant python test/quantization/test_qat.py -k test_qat_8da4w_quantizer_disable_fake_quant_backward Reviewers: jerryzh168, cpuhrsch Subscribers: jerryzh168, cpuhrsch, supriyar
andrewor14
force-pushed
the
8da4w_qat_enable
branch
from
fbc5742
to
56afc27
Compare
|
@andrewor14 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@andrewor14 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@andrewor14 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@andrewor14 maybe you can unlink the diff |
How can I do that? |
|
@andrewor14 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
Summary: This feature helps with model convergence during QAT. The user can disable observation/fake quant for the first N steps and renable them later, allowing the activation and weight values to stabilize before applying quantization. Test Plan: python test/quantization/test_qat.py -k test_qat_8da4w_quantizer_disable_fake_quant python test/quantization/test_qat.py -k test_qat_8da4w_quantizer_disable_fake_quant_backward Reviewers: jerryzh168, cpuhrsch Subscribers: jerryzh168, cpuhrsch, supriyar
Summary: This feature helps with model convergence during QAT. The user can disable observation/fake quant for the first N steps and renable them later, allowing the activation and weight values to stabilize before applying quantization.
Test Plan:
python test/quantization/test_qat.py -k test_qat_8da4w_quantizer_disable_fake_quant
python test/quantization/test_qat.py -k test_qat_8da4w_quantizer_disable_fake_quant_backward
Reviewers: jerryzh168, cpuhrsch
Subscribers: jerryzh168, cpuhrsch, supriyar