moyix

#!/usr/bin/env python

import os
import sys
import subprocess as sp
import tempfile
import hashlib
script_dir = os.path.dirname(os.path.realpath(__file__))
sys.path.append(script_dir)
from fast_check_for_ffast_math import check_file

moyix

import sys
import mmap
from elftools.elf.elffile import ELFFile, ELFError
import struct

set_fast_math_code = bytes.fromhex('0fae5c24fc814c24fc408000000fae5424fcc3')

def load_bytes_from_elf(bindata, elf, vaddr, size):
    try:
        paddr = next(iter(elf.address_offsets(vaddr)))

moyix

$ objdump -s -j .init_array ./jaxlib/xla_extension.so | sed -e '1,/Contents/ d' | cut -c 10-44 | xxd -r -p | od -A none -w8 -t x8 --endian=little | addr2line -a -f -e ./jaxlib/xla_extension.so | paste -sd '  \n' | c++filt
0x000000000084c5e0 __cpu_indicator_init /dt9-src/libgcc/config/i386/cpuinfo.c:434
0x000000000084ca20 frame_dummy crtstuff.c:?
0x000000000079c440 _GLOBAL__sub_I_xla.cc xla.cc:?
0x000000000079c540 _GLOBAL__sub_I_dlpack.cc dlpack.cc:?
0x000000000079c5f0 _GLOBAL__sub_I_mlir.cc mlir.cc:?
0x000000000079c620 _GLOBAL__sub_I_ops.cc ops.cc:?
0x000000000079c650 _GLOBAL__sub_I_approx_topk.cc approx_topk.cc:?
0x000000000079c680 _GLOBAL__sub_I_approx_topk_shape.cc approx_topk_shape.cc:?
0x000000000079c6b0 _GLOBAL__sub_I_lu_decomposition.cc lu_decomposition.cc:?

moyix

#!/usr/bin/env python

import subprocess
import re
import sys

def get_init_array(filename):
    # Call objdump -s -j .init_array <filename> to get the contents of the .init_array section
    try:
        objdump_output = subprocess.check_output(['objdump', '-s', '-j', '.init_array', filename], stderr=subprocess.STDOUT)

moyix

(sfcodegen) moyix@isabella:~$ python load_codegen_with_longer_context.py 
vocab_file vocab.json
merges_file merges.txt
tokenizer_file tokenizer.json
added_tokens_file added_tokens.json
special_tokens_map_file special_tokens_map.json
tokenizer_config_file tokenizer_config.json
Partial prompt from /usr/include/stdlib.h:

[...] restrict __nptr,

moyix

#!/usr/bin/env python

import torch
from transformers import CodeGenConfig, CodeGenForCausalLM, CodeGenTokenizer
from transformers.utils.hub import cached_file

NEW_SIZE = 4096

cg_config = CodeGenConfig.from_pretrained('Salesforce/codegen-350M-mono')
cg_config.n_ctx = NEW_SIZE

moyix

#!/usr/bin/env python

import time
import random
import string
from flask import Flask, request
import numpy as np
import tritonclient.grpc as client_util
import json
from tritonclient.utils import np_to_triton_dtype

moyix

# Copyright (c) 2021-2022, NVIDIA CORPORATION.  All rights reserved.
# Modified by Brendan Dolan-Gavitt, 2022
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software

moyix

Using Linear Algebra to Convert a Large Code Model

Background

The SalesForce CodeGen models are a family of large language models trained on a large amount of natural language data and then fine-tuned on specialized datasets of code. Models of size 350M, 2B, 6B, and 16B parameters are provided in three flavors:

• nl, the base model trained on The Pile, a large natural language dataset compiled by EleutherAI
• multi, which is fine-tuned from the nl model on a dataset of code in multiple languages, scraped from GitHub, and
• mono, which is fine-tuned from the multi model on Python code only.

moyix

#!/usr/bin/env python

import argparse
import torch
from transformers import GPTJForCausalLM, GPTJConfig
# Note: these need the git version of Transformers as of 7/22/2022
from transformers import CodeGenTokenizer, CodeGenForCausalLM
from transformers import CODEGEN_PRETRAINED_MODEL_ARCHIVE_LIST

parser = argparse.ArgumentParser('Convert SalesForce CodeGen model to GPT-J')