Exported to Compiler Graph FXIR
Submodules
dmx.compressor.dmx.nn.Linear
to_compiler_graph
opcode name target args kwargs
------------- ---------------------- -------------------------- ------------------------------------------------------------ --------
placeholder _input _input () {}
get_attr input_cast_scale input_cast.scale () {}
get_attr input_cast_zero_point input_cast.zero_point () {}
call_function quantize dmx.quantize (_input, input_cast_scale, input_cast_zero_point, 'SAME') {}
call_function dequantize dmx.dequantize (quantize, input_cast_scale, input_cast_zero_point) {}
get_attr _weight _weight () {}
get_attr weight_scale weight_scale () {}
get_attr weight_zero_point weight_zero_point () {}
call_function quantize_1 dmx.quantize (_weight, weight_scale, weight_zero_point, 'SAME') {}
call_function dequantize_1 dmx.dequantize (quantize_1, weight_scale, weight_zero_point) {}
get_attr _bias _bias () {}
get_attr bias_cast_scale bias_cast.scale () {}
get_attr bias_cast_zero_point bias_cast.zero_point () {}
call_function quantize_2 dmx.quantize (_bias, bias_cast_scale, bias_cast_zero_point, 'SAME') {}
call_function dequantize_2 dmx.dequantize (quantize_2, bias_cast_scale, bias_cast_zero_point) {}
call_function _output <built-in function linear> (dequantize, dequantize_1, dequantize_2) {}
get_attr output_cast_scale output_cast.scale () {}
get_attr output_cast_zero_point output_cast.zero_point () {}
call_function quantize_3 dmx.quantize (_output, output_cast_scale, output_cast_zero_point, 'SAME') {}
call_function dequantize_3 dmx.dequantize (quantize_3, output_cast_scale, output_cast_zero_point) {}
output output output (dequantize_3,) {}
dmx.compressor.dmx.nn.ResAdd
to_compiler_graph
opcode name target args kwargs
------------- ------------------------ ------------------------------------------------------ ----------------------------------------------------------------- --------
placeholder _input _input () {}
get_attr input_cast_scale input_cast.scale () {}
get_attr input_cast_zero_point input_cast.zero_point () {}
call_function quantize dmx.quantize (_input, input_cast_scale, input_cast_zero_point, 'SAME') {}
call_function dequantize dmx.dequantize (quantize, input_cast_scale, input_cast_zero_point) {}
placeholder residual residual () {}
get_attr residual_cast_scale residual_cast.scale () {}
get_attr residual_cast_zero_point residual_cast.zero_point () {}
call_function quantize_1 dmx.quantize (residual, residual_cast_scale, residual_cast_zero_point, 'SAME') {}
call_function dequantize_1 dmx.dequantize (quantize_1, residual_cast_scale, residual_cast_zero_point) {}
call_function output <built-in method add of type object at 0x7f9ef8059840> (dequantize, dequantize_1) {}
get_attr output_cast_scale output_cast.scale () {}
get_attr output_cast_zero_point output_cast.zero_point () {}
call_function quantize_2 dmx.quantize (output, output_cast_scale, output_cast_zero_point, 'SAME') {}
call_function dequantize_2 dmx.dequantize (quantize_2, output_cast_scale, output_cast_zero_point) {}
output output_1 output (dequantize_2,) {}
dmx.compressor.dmx.nn.ActActMatMul
to_compiler_graph
opcode name target args kwargs
------------- -------------------------- --------------------------------------------------------- ----------------------------------------------------------------------- --------
placeholder _input _input () {}
get_attr input_cast_scale input_cast.scale () {}
get_attr input_cast_zero_point input_cast.zero_point () {}
call_function quantize dmx.quantize (_input, input_cast_scale, input_cast_zero_point, 'SAME') {}
call_function dequantize dmx.dequantize (quantize, input_cast_scale, input_cast_zero_point) {}
placeholder multiplier multiplier () {}
get_attr multiplier_cast_scale multiplier_cast.scale () {}
get_attr multiplier_cast_zero_point multiplier_cast.zero_point () {}
call_function quantize_1 dmx.quantize (multiplier, multiplier_cast_scale, multiplier_cast_zero_point, 'SAME') {}
call_function dequantize_1 dmx.dequantize (quantize_1, multiplier_cast_scale, multiplier_cast_zero_point) {}
call_function output <built-in method matmul of type object at 0x7f9ef8059840> (dequantize, dequantize_1) {}
get_attr output_cast_scale output_cast.scale () {}
get_attr output_cast_zero_point output_cast.zero_point () {}
call_function quantize_2 dmx.quantize (output, output_cast_scale, output_cast_zero_point, 'SAME') {}
call_function dequantize_2 dmx.dequantize (quantize_2, output_cast_scale, output_cast_zero_point) {}
output output_1 output (dequantize_2,) {}
dmx.compressor.dmx.nn.Embedding
to_compiler_graph
opcode name target args kwargs
------------- --------- -------------------------------------- ----------------- -------------------------------------------------------------------------------------------------------
placeholder input_1 input () {}
get_attr weight weight () {}
call_function embedding <function embedding at 0x7f9e211aadd0> (input_1, weight) {'padding_idx': None, 'max_norm': None, 'norm_type': 2.0, 'scale_grad_by_freq': False, 'sparse': False}
output output output (embedding,) {}
dmx.compressor.dmx.nn.SoftMax
to_compiler_graph
opcode name target args kwargs
------------- ------- ------------------------------------ ---------- --------
placeholder _input _input () {}
call_function softmax <function softmax at 0x7f9e211aa7a0> (_input,) {}
output output output (softmax,) {}
dmx.compressor.dmx.nn.LayerNorm
to_compiler_graph
opcode name target args kwargs
------------- -------------------- --------------------------------------- ------------------------------------------------------- --------
placeholder _input _input () {}
get_attr _weight _weight () {}
get_attr weight_scale weight_scale () {}
get_attr weight_zero_point weight_zero_point () {}
call_function quantize dmx.quantize (_weight, weight_scale, weight_zero_point, 'SAME') {}
call_function dequantize dmx.dequantize (quantize, weight_scale, weight_zero_point) {}
get_attr _bias _bias () {}
get_attr bias_cast_scale bias_cast.scale () {}
get_attr bias_cast_zero_point bias_cast.zero_point () {}
call_function quantize_1 dmx.quantize (_bias, bias_cast_scale, bias_cast_zero_point, 'SAME') {}
call_function dequantize_1 dmx.dequantize (quantize_1, bias_cast_scale, bias_cast_zero_point) {}
get_attr normalized_shape normalized_shape () {}
get_attr eps eps () {}
call_function ln <function layer_norm at 0x7f9e211ab1c0> (_input, normalized_shape, quantize, dequantize_1, eps) {}
output output output (ln,) {}
dmx.compressor.dmx.nn.Dropout
to_compiler_graph
opcode name target args kwargs
------------- ------- ------------------------------------ ---------- ----------------------------------------------
placeholder input_1 input () {}
call_function dropout <function dropout at 0x7f9e211a9bd0> (input_1,) {'p': 0.5, 'training': True, 'inplace': False}
output output output (dropout,) {}
dmx.compressor.dmx.nn.GELU
to_compiler_graph
opcode name target args kwargs
------------- ------- ------------------------ ---------- -----------------------
placeholder input_1 input () {}
call_function gelu <built-in function gelu> (input_1,) {'approximate': 'none'}
output output output (gelu,) {}