aimet icon indicating copy to clipboard operation
aimet copied to clipboard

Not loading encodings for quantizer: ****** as it is disabled in Tensorflow framework

Open tensor1to5 opened this issue 1 year ago • 1 comments

Hi, I have tried the following code that use the function sim.load_encoding_to_sim to achieve mixed precision quantization of operators :

from aimet_tensorflow.keras.quantsim import QuantizationSimModel
from aimet_common.defs import QuantScheme

sim = QuantizationSimModel(model=model,
                           quant_scheme=QuantScheme.post_training_tf,
                           rounding_mode="nearest",
                           default_output_bw=16,
                           default_param_bw=8)
sim.load_encoding_to_sim("./model_after_qat.encodings")
sim.compute_encodings(forward_pass_callback=pass_calibration_data,
                      forward_pass_callback_args=1000)

and meet a problem,but this issue is not present in the Torch framework : Not loading encodings for quantizer: ****** as it is disabled

tensor1to5 avatar Sep 28 '23 04:09 tensor1to5

@quic-ernst, could you take a look at this? Thanks

quic-mangal avatar Sep 28 '23 17:09 quic-mangal