Lower quant/dequant torch op to StableHLO #5763
lsy323
merged
23 commits into
master
from
lsiyuan/quant-dequant-dispatch
JackCaoG
commented
on 2023-11-02
Add q/dq custom op in torch_xla
ea312792
Add stablehlo converter change to support quantized tensor. Update
c6bc19eb
Add per-channel quant, add resnet test
79632c23
add stablehlo patch for serializing qdq into bytecode
108992b0
add support of exporting qdq to tf.saved_model
1adc299b
make xla quant/dequant op compatible with decomp table
3efb881f
format
9a6bda1e
untrack some tests
23b2aa5e
move resnet test under stablehlo test folder
7f620923
add patch for stablehlo.uniform_quantize/dequantize bytecode serializ…
a5122419
rename files
3d00629f
use python dispatcher for qdq
341c6668
fix compile
c779522d
fix linter
7f79a494
add per tensor/channel tests
fbb68e07
format
163de141
have all scalers have positive values
c329b63c
add comment for STABLEHLO_BYTECODE_FROM_PRETTYPRINT
9177ae0f
JackCaoG
commented
on 2023-11-27
JackCaoG
commented
on 2023-11-27
JackCaoG
commented
on 2023-11-27
JackCaoG
commented
on 2023-11-27
JackCaoG
commented
on 2023-11-27
JackCaoG
commented
on 2023-11-27
JackCaoG
commented
on 2023-11-27
do not udpate save_torch_module_as_tf_saved_model signature to take g…
b852cc96
add comment for QuantParams struct
5d28a73d
add new lines to the end of header file
2253b2bd
update test
1560e724
add assertions, use int for zp
3e5462d8
JackCaoG
approved these changes
on 2023-11-28
miladm
commented
on 2023-12-01
Login to write a write a comment.
Login via GitHub