issues
search
cjg91
/
trans-fat
An FPGA Accelerator for Transformer Inference
72
stars
12
forks
source link
3.29 Meeting Minutes
#5
Open
cjg91
opened
2 years ago
cjg91
commented
2 years ago
Colman Tasks:
Input/output definition of each stage (with unnecessary requantization removed)
Host code interface
Dan:
int8 -> int32 systolic matrix multiplier
initialize accumulator with bias vector
Owen:
c++ softmax implementation following
tensor_quant_softmax
or equivalent quantized implementation
kernel softmax implementation
Colman Tasks:
Dan:
Owen:
tensor_quant_softmax
or equivalent quantized implementation