allocated weigths

[Gabriel-MarcoSilva]

Hello everyone, I have a question. I used hls4ml, it generated the verilogs files, but the question is, how are the weights allocated? In this case: identify where the allocated weights are for each layer.

[bo3z]

Depends on the strategy.

With Latency, these are "embedded" as constants straight into the Verilog files.

In Resource, there should be .dat files which correspond to the contents of the weight matrix stored in BRAM.

[Gabriel-MarcoSilva]

I use this configurations:

config = hls4ml.utils.config_from_keras_model(model, granularity='name', backend='Vitis')
config['Model']['IOType'] = 'io_stream' # Usar streams para entradas/saídas
config['Model']['Precision'] = 'ap_fixed<16,10>'
config['Model']['ReuseFactor'] = 2
config['Model']['Strategy'] = 'Latency'