xla
a3b0c6ec - Lower quant/dequant torch op to StableHLO (#5763)

Commit

1 year ago

Lower quant/dequant torch op to StableHLO (#5763) (de)quantize_per_tensor/channel ops from PT2E quantization workflow are lowered to stablehlo uniform_dequantize/quantize. --------- Co-authored-by: Siyuan Liu <lsiyuan@google.coim>

References

#5763 - Lower quant/dequant torch op to StableHLO

Author

lsy323

Parents

4ac255ae

Files24

WORKSPACE
openxla_patches
- quant_dequant_converter.diff
- stablehlo_quant_seralization.diff
test/stablehlo
- test_pt2e_qdq.py
torch_xla
- csrc
  - BUILD
  - init_python_bindings.cpp
  - ir_dump_util.cpp
  - ops
    - dequant_tensor.cpp
    - dequant_tensor.h
    - quant_tensor.cpp
    - quant_tensor.h
    - xla_ops.cpp
    - xla_ops.h
  - quant_util.cpp
  - quant_util.h
  - runtime
    - BUILD
    - pjrt_computation_client.cc
    - stablehlo_helper.cc
    - stablehlo_helper.h
  - tensor_methods.cpp
  - tensor_methods.h
  - xla_graph_executor.cpp
- experimental
  - quantized.py
- stablehlo.py

xla a3b0c6ec - Lower quant/dequant torch op to StableHLO (#5763)

xla
a3b0c6ec - Lower quant/dequant torch op to StableHLO (#5763)