mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-06 12:20:52 +01:00
6f63126b5c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/54860
Currently we insert a quantize_per_tensor op when we encounter the quantizable input,
so if it has multiple uses and not all are quantizable then we need to add a dequantize op
before these ops.
In this pass - For a sequence of quantize_per_tensor - dequantize, we combine them
since it is a no-op.
[internal only][pyper]
Before this change we had redundant dequantize nodes in the graph
Example 1x inline_cvr graph https://www.internalfb.com/intern/everpaste/?handle=GODBxAlUMzGHD6 (
|
||
|---|---|---|
| .. | ||
| fx | ||
| ns | ||
| __init__.py | ||
| _correct_bias.py | ||
| _equalize.py | ||
| _learnable_fake_quantize.py | ||
| _numeric_suite_fx.py | ||
| _numeric_suite.py | ||
| fake_quantize.py | ||
| fuse_modules.py | ||
| fuser_method_mappings.py | ||
| observer.py | ||
| qconfig.py | ||
| quant_type.py | ||
| quantization_mappings.py | ||
| quantize_fx.py | ||
| quantize_jit.py | ||
| quantize.py | ||
| stubs.py | ||
| utils.py | ||