sourabh2k15

#@title FFT  Layer

"""Flax layer to perform preprocessing on librispeech audio inputs.

This layer computes windowed short time fourier transform over audio signals
then converts it to mel scale and finally takes a logarithm of resulting
mel spectrograms and normalizes it to be used in speech recognition models.

This code is based on lingvo's librispeech preprocessing code here:
https://github.com/tensorflow/lingvo/blob/master/lingvo/tasks/asr/frontend.py

sourabh2k15

JAX training steps progress : 

0) loss = 32.684505462646484 grad_norm = 46.189605712890625
1) loss = 32.684505462646484 grad_norm = 46.189605712890625
2) loss = 32.2853889465332 grad_norm = 47.51691818237305
3) loss = 31.46957778930664 grad_norm = 49.439979553222656
4) loss = 30.260791778564453 grad_norm = 53.63009262084961
5) loss = 28.776248931884766 grad_norm = 58.055015563964844
6) loss = 27.165119171142578 grad_norm = 73.13194274902344
7) loss = 25.350372314453125 grad_norm = 83.24612426757812

sourabh2k15

global_step,test/ctc_loss,test/wer,train/ctc_loss,train/wer,validation/ctc_loss,validation/wer,score
1,30.24002,1.1590399418146555,32.369118,1.0535290080768622,30.846983,1.1507724619108644,
528,5.9877963,0.8947815019092066,6.026844,0.9461873981817226,6.1750135,0.8924567737032111,
1062,6.1004605,0.8944582499949492,6.3094215,0.9392578586622548,6.2672987,0.8986089921924078,
1598,4.909699,0.8416267652585006,4.9721985,0.8754789272030651,5.293022,0.852067526555387,
2124,3.8002286,0.7468331414025092,3.7078342,0.7777719500681842,4.0218964,0.7641824646666786,
2651,2.7804344,0.6221387154777057,2.787846,0.6536303051560856,3.2950375,0.6540194987648268,
3176,1.7282625,0.4829585631452411,1.7336979,0.5041396285522488,1.9899195,0.5062177017329219,
3703,1.1341795,0.3567286906277148,1.0678067,0.3458033573141487,1.6042109,0.4285113463894527,
4227,0.7807562,0.2609047012950279,0.6385119,0.2235402865506085,1.0614083,0.315793368857312,

sourabh2k15

sudo apt-get install linux-headers-$(uname -r)
sudo apt-get update
sudo apt-get upgrade 
sudo apt-get install gcc
sudo apt-get install libxml2
sudo apt-get install make 
sudo apt-get install python3-pip
sudo apt-get install git
sudo apt-get install tmux

sourabh2k15

Thread 767 (Thread 0x7f8a0933a700 (LWP 54659)):
#0  futex_wait_cancelable (private=0, expected=0, futex_word=0x7f868797a358) at ../sysdeps/unix/sysv/linux/futex-internal.h:88
#1  __pthread_cond_wait_common (abstime=0x0, mutex=0x7f868797a360, cond=0x7f868797a330) at pthread_cond_wait.c:502
#2  __pthread_cond_wait (cond=0x7f868797a330, mutex=0x7f868797a360) at pthread_cond_wait.c:655
#3  0x00007fb6a13d9f18 in persistentThread(void*) () from /home/smedapati_google_com/mlcommons2/lib/python3.7/site-packages/jaxlib/xla_extension.so
#4  0x00007fb780c87fa3 in start_thread (arg=<optimized out>) at pthread_create.c:486
#5  0x00007fb780a1feff in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

sourabh2k15

class Solution {
public:
    vector<vector<string>> solveNQueens(int n) {
        vector<vector<string>> solutions;
        vector<string> board(n, string(n, '.'));
        
        solveBoard(solutions, board, 0, n);   
        return solutions;
    }
    

sourabh2k15

class Solution {
private:
    vector<vector<int>> powerset;
    vector<int> subset;
public:
    vector<vector<int>> subsets(vector<int>& nums) {        
        backtrack(nums, 0);
        return powerset;
    }

sourabh2k15

class Solution {
private:
    vector<int> combination;
    vector<vector<int>> combinations;
public:
    vector<vector<int>> combinationSum2(vector<int>& candidates, int target) {
        sort(candidates.begin(), candidates.end());
        
        explore(candidates, 0, target);
        return combinations;

sourabh2k15

class Solution {
private:
    vector<int> combination;
    vector<vector<int>> combinations;

public:
    vector<vector<int>> combinationSum(vector<int>& candidates, int target) {
        explore(candidates, 0, target);
        return combinations;
    }

sourabh2k15

class Solution {
private:
    vector<vector<int>> powerset;
    vector<int> subset;
public:
    vector<vector<int>> subsetsWithDup(vector<int>& nums) {
        sort(nums.begin(), nums.end()); //change 1
        
        backtrack(nums, 0);
        return powerset;