My question:
Hi,I am currently using the following INT8 quantization configuration:
INT8_WEIGHT_ONLY_CFG = { "quant_cfg": { "*weight_quantizer": {"num_bits": 8, "axis": 0}, "*input_quantizer": {"enable": False}, **_default_disabled_quantizer_cfg, }, "algorithm": "max", }
I would like to use the histogram calibration algorithm instead of max. Could you please advise on how to correctly modify this configuration to enable histogram-based calibration?
Specifically, should I simply change the "algorithm" field, or are there additional parameters required within "quant_cfg"?
Thanks!