twobob

make win64 && make winclang && make wingcc

for /L %i in (1,1,3) do @(for %x in (runmingw.exe rungcc.exe run.exe) do (set OMP_NUM_THREADS=4 && %x ../out/model110M.bin 0 0 "Once upon a time" 0))

for /f "tokens=4" %i in ('chcp') do @(chcp 65001 && @echo off && for /L %j in (1,1,3) do @(for %x in (runmingw.exe rungcc.exe run.exe) do @(set OMP_NUM_THREADS=4 && %x ../out/model110M.bin 0 0 "And away they went" 0)) && @echo on && chcp %i)

twobob

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdint.h>
#include <direct.h>
#include <time.h>

static inline void process_triplet(const char *input, size_t i, uint32_t *triplet) {
    *triplet = (input[i] << 16) | (i + 1 < strlen(input) ? input[i + 1] << 8 : 0) | (i + 2 < strlen(input) ? input[i + 2] : 0);
}

twobob

/*
Inference for Llama-2 Transformer model in pure C.
This version will spit out story blocks as fast as possible to a folder called inbox
Metrics are shown per story, no doubt this could be faster.

Output using -03 and no -fopenmp, with token-by-token reporting on the test machine gave 6-8 tok/s second.
Compiling as outlined below and foregoing constant screen output nets between 80-330 tok/s on the same machine.
So between 10 - 55 times faster.

Example compile: (see README for more details)

twobob

/*
Inference for Llama-2 Transformer model in pure C.

Example compile: (see README for more details)
$ gcc -O3 -o run run.c -lm

Then run with:
$ ./run
*/

twobob

 	#!/usr/bin/env python
	
	# Based on: https://github.com/oobabooga/text-generation-webui/blob/main/convert-to-torch.py
	# License: GNU Affero General Public License v3.0
	#
	#
	# This script converts a transformers model using a custom shard size.
	#
	# Load a model from a directory and shard it into 2GB chunks:
	# python reshard-causallm-model.py --src-model gpt-j-6B --out-path gpt-j-6B-sharded --torch_dtype float16 --max-shard-size 2GB

twobob

import json

def is_valid_jsonl(file_path):
    with open(file_path, 'r', encoding='utf-8') as f:
        for line_number, line in enumerate(f, start=1):
            try:
                json.loads(line)
            except json.JSONDecodeError:
                print(f'Invalid JSON on line {line_number}: {line}')
                return False

twobob

import os
import heapq
import shutil
import argparse

def get_directory_size(directory):
    """
    Calculate the size of the directory.

    Parameters:

twobob

import argparse
import os
import sys
import random
import codecs
import json
from bs4 import BeautifulSoup
from multiprocessing import Pool
import colorama
from colorama import Fore

twobob

!wget https://model-server.zqevans2.workers.dev/jmann-small-190k.ckpt -O /content/drive/MyDrive/AI/models/jmann-small-190k.ckpt

twobob

du -cBM --max-depth=1 2> >(grep -v 'Permission denied') | grep -v 'cannot access' | sort -n