rachtsingh

"""
This script exports the AutoGPT-Q Llama 2 weights in llama2rs.bin format.
"""
import pathlib
import click
import struct
import torch
from torch import nn
from auto_gptq import AutoGPTQForCausalLM
from auto_gptq.nn_modules import qlinear

rachtsingh

#!/usr/local/bin/python3
"""
This script should idempotently grab the stuff that's newly added to Pocket to pocket.md

You'll need to install https://airtable-python-wrapper.readthedocs.io, i.e. via:

    $ pip3 install airtable-python-wrapper

and, of course, sign up for an Airtable account (free seems to work fine).
"""

rachtsingh

#include <stdio.h>
#include <stdlib.h>

// void * return type means that it returns a naked pointer, which is just an address.
void *foo(void) {
  long *bar_ptr = malloc(sizeof(long));
  *bar_ptr = 5;
  printf("The data is held at address: %p, holds %ld bytes\n", bar_ptr, sizeof(long));
  free(bar_ptr);
  return (void *) bar_ptr;

rachtsingh

#include <stdio.h>

// void * return type means that it returns a naked pointer, which is just an address.
void *foo(void) {
  long bar = 5;
  printf("bar's address is: %p, holds %ld bytes\n", &bar, sizeof(long));
  long *ret = &bar;
  return (void *) ret; // ret is a pointer to a long, but we just turn it into a regular address
  // note that the previous line doesn't *do* anything, it just gets the compiler off our back
}

rachtsingh

import numpy as np 
from matplotlib import pyplot as plt

def sample_gamma(alpha):
    total = 0
    scale = 1.

    if (alpha < 1.0):
        scale *= (1 - np.random.uniform() ** (1.0 / alpha))
        total += 1

rachtsingh

#include "ATen/NativeFunctions.h"
#include "ATen/Dispatch.h"
#include "ATen/cuda/CUDAApplyUtils.cuh"
#include <curand.h>
#include <curand_kernel.h>
#include <curand_philox4x32_x.h>
#include <utility>
#include <functional>
#include <nvfunctional>

rachtsingh

Keybase proof

I hereby claim:

• I am rachtsingh on github.
• I am rachitsingh (https://keybase.io/rachitsingh) on keybase.
• I have a public key ASAtaQuPwR-0aYEy7PkgjheFZh-b4C2ZfWG7ssWrHRDi8Qo

To claim this, I am signing this object:

rachtsingh

[01/30/17 13:57:47 INFO] Using GPU(s): 1	
[01/30/17 13:57:47 INFO] Loading data from '../data/translate-train.t7'...	
[01/30/17 13:57:51 INFO]  * vocabulary size: source = 50004; target = 50004	
[01/30/17 13:57:51 INFO]  * additional features: source = 0; target = 0	
[01/30/17 13:57:51 INFO]  * maximum sequence length: source = 50; target = 51	
[01/30/17 13:57:51 INFO]  * number of training sentences: 100000	
[01/30/17 13:57:51 INFO]  * maximum batch size: 64	
[01/30/17 13:57:51 INFO] Building model...	
[01/30/17 13:57:55 INFO]  * using input feeding	
[01/30/17 13:57:56 INFO] Initializing parameters...	

rachtsingh

[01/30/17 12:34:39 INFO] Using GPU(s): 1	
[01/30/17 12:34:39 INFO] Loading data from '../data/translate-train.t7'...	
[01/30/17 12:34:44 INFO]  * vocabulary size: source = 50004; target = 50004	
[01/30/17 12:34:44 INFO]  * additional features: source = 0; target = 0	
[01/30/17 12:34:44 INFO]  * maximum sequence length: source = 50; target = 51	
[01/30/17 12:34:44 INFO]  * number of training sentences: 100000	
[01/30/17 12:34:44 INFO]  * maximum batch size: 64	
[01/30/17 12:34:44 INFO] Building model...	
[01/30/17 12:34:48 INFO]  * using input feeding	
[01/30/17 12:34:48 INFO] Initializing parameters...	

rachtsingh

Loading data from 'data/small-train.t7'...	
 * vocabulary size: source = 50004; target = 50004	
 * additional features: source = 0; target = 0	
 * maximum sequence length: source = 50; target = 51	
 * number of training sentences: 100000	
 * maximum batch size: 64	
Building model...	
 * using input feeding	
Initializing parameters...	
 * number of parameters: 84834004