About

Anam Nikhilesh Kumar Reddy is a Software Engineer at Kinara, Inc. with 1.78 years of relevant experience in the field. Anam is part of the Neural Networks and Deep Learning Compiler Team at Kinara, Inc. and is responsible for various projects related to MxNet, Tensorflow, Pytorch, and Caffe, among others. Anam has extensive experience in Ops Validation, Pattern Matchers, ConvNext, Fast Reid, and Mask Softmax for Transformers Refiner1. Anam has written scripts to generate randomized test models with constraints and seeds from the ops.pbtxt version for MxNet and Tensorflow frameworks, fixed numerous bugs, and developed the code for the Converter stack by using the generated test cases by validating the ops conversion to caffe framework. Anam has also implemented a pattern matcher framework using networkx library for MxNet to match and replace subgraphs in the model and implemented Fully Connected pattern matcher and Pad fusion to Convolution and MAX Pooling layers for a given refiner model. Anam has used the 8-bit quantization format to implement fixed point approximation of Gaussian Error Linear Unit activation for ConvNext model and implemented unit width, height, and channel kernels for signed and unsigned GELU. Anam has also converted the convNext model from pytorch to onnx and implemented pattern matchers for LayerNorm. Anam has fused the pattern of matmul followed by a large division scalar to a single batchdot with 16-bit quantizes scalar multiplication for Fast Reid, and improved SNR by 3x after modifying the datapoints using lapacian distribution. Anam has implemented a mxnet pattern matcher for sequence mask followed by softmax to improve the performance of the model, and improved the SNR of the fused layer by 3.5 times. Anam has used various languages such as Python, C, C++, and technologies such as Gitlab, Jenkins, and tools such as Mxnet, Onnx, Pytorch, Tensorflow, Caffe for the projects. Anam has also used debugging tools such as pdb, gdb, and valgrind tools, and has experience working with Protobufs, Make, CMake, and Build Systems. Anam has previously worked as a Software Development Intern at aerchain and rite infotech, where Anam built a real-time supplier dashboard analytics and reports based on required metrics using NodeJS, SequelizeJS, PostgreSQL, Amplitude Analytics, and