jiqing-feng 2f4a7177f0 Fix training gradient underflow in quantization tests (#13539) ...

* Fix training gradient underflow in quantization tests

Change autocast dtype from float16 to bfloat16 in _test_quantization_training.
Float16's limited dynamic range causes gradients to underflow to zero when
passing through quantized tensor subclass operations.

* fix autocast dtype check

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

---------

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

2026-05-18 15:20:07 +05:30

..

__init__.py

[tests] add attention backend tests. (#13174)

2026-05-12 17:00:00 +09:00

attention.py

[tests] add attention backend tests. (#13174)

2026-05-12 17:00:00 +09:00

cache.py

Refactor Model Tests (#12822)

2026-02-02 18:51:44 +05:30

common.py

[CI] Fix BnB tests (#13481)

2026-04-22 13:19:32 +05:30

compile.py

[tests] accept recompile_limit from the user in tests (#13150)

2026-02-16 14:48:21 +05:30

ip_adapter.py

Refactor Model Tests (#12822)

2026-02-02 18:51:44 +05:30

lora.py

[tests] fix lora logging tests for models. (#13318)

2026-03-24 15:48:03 +05:30

memory.py

[tests] fix group offloading with disk tests (#13491)

2026-04-22 11:41:37 -03:00

parallelism.py

add SP support for flash_varlen_hub backend (#13479)

2026-05-10 08:41:08 +09:00

quantization.py

Fix training gradient underflow in quantization tests (#13539)

2026-05-18 15:20:07 +05:30

single_file.py

Refactor Model Tests (#12822)

2026-02-02 18:51:44 +05:30

training.py

Refactor Model Tests (#12822)

2026-02-02 18:51:44 +05:30

utils.py

[tests] add attention backend tests. (#13174)

2026-05-12 17:00:00 +09:00