Amin Ghasemazar

icon indicating a website link ece.ubc.ca/~aming icon indicating an email [email protected]

icon company UBC icon location Vancouver icon location I am a last year PhD candidate in ECE at the UBC, and RA in the System-on-Chip Lab. - ML Compression, -Cache Compression -Approximate Comp

Results 5 issues of Amin Ghasemazar

Higher than 8-bit Quantization not working properly!?

Thanks for this great framework! I was wondering if there is an explicit 'no' or a limitation for quantizing weights and/or activations to higher than 8 bits using asymmetric methods?...

Update prepare_model_quant.html

Change the 2nd relu function name to `relu2`.

Issue with other than 8-bit activations

I believe the code here [Here](https://github.com/jakc4103/DFQ/blob/6f15805cfdbf2769275defd54728df0a5d30dbc6/utils/quantize.py#L366) should be repaced with `graph[idx].quant = QuantMeasure(num_bits=bits_activation)`. As it is missing the proper argument name, it will always add a QuantMeasure module with 8-bit...

‘is_layer’ member issue

1 comment

Hi, I followed the readme file and I keep getting this message while trying to compile the benchmark. ``` btree_impl.h:213:7: error: ‘struct btree::leaf_node’ has no member named ‘is_layer’ ``` Any...

2x accLat in timing cache

Hey, to the best of my knowledge, ZSim models the timing (conventional) cache as if we need one accLat to access TAG and DATA arrays. however in newer designs, we...