assert isinstance(quantization_spec, QuantizationSpec)

[barbecacov]

🐛 Describe the bug

Under the lastest executorch (0.8.0a0)， and using the quantize function like this:
def quantize(model, example_inputs):
"""This is the official recommended flow for quantization in pytorch 2.0 export"""
print(f"Original model: {model}")
quantizer = XNNPACKQuantizer()
# if we set is_per_channel to True, we also need to add out_variant of quantize_per_channel/dequantize_per_channel
operator_config = get_symmetric_quantization_config(is_dynamic=True, is_per_channel=True)
quantizer.set_global(operator_config)
m = prepare_pt2e(model, quantizer)

then i got
torch/ao/quantization/fx/prepare.py", line 182, in _create_obs_or_fq_from_qspec
assert isinstance(quantization_spec, QuantizationSpec)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

but when i using executorch 0.6.0, it is fine

Versions

0.8.0a0

cc @digantdesai @mcr229 @cbilgin @kimishpatel @jerryzh168

[swolchok]

This looks like intersection of quantization and XNNPACK if I understand correctly. I think the right triage is to @jerryzh168, but CC @digantdesai .

@barbecacov thank you for reporting a bug! It would help us address your issue more quickly if you could also do the following:

1. Provide a complete program that we can just copy/paste and run to reproduce the issue, and/or
2. Include the complete backtrace ending with the assertion failure you pasted above, not just the last line.

[jerryzh168]

this could be related to the migration we did recently, can you post full examples including the import path for prepare_pt2e and convert_pt2e, quantizer etc.

cc @metascroy

[barbecacov]

@jerryzh168 Thank you so much, and Sorry for that i used toch.ao to import like v0.6.0 in executorch version 0.8.0
from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
and when i use
from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e

there is no error. thank you so much