nreHieW

"""
The 2024 Transformer (the Noam Transformer):
- RMSNorm
- GQA or some combination 
- Sliding window attention 
- Swiglu
- RoPE (Rotary Positional Embedding)

LLM Arches:
                hidden | MLP mult.   |  n_layers | rope_theta |  GQA Group Size  | GLU Act. |  ops 

soulitzer

import torch
import functools
from torch.utils._python_dispatch import TorchDispatchMode
import torch.utils._pytree as pytree
from torch.utils.weak import WeakTensorKeyDictionary


class RecomputableTensor(torch.Tensor):
    @staticmethod
    def __new__(cls, t, func, args):

mlazos

`--> TORCH_LOGS="output_code" python optim_repro.py
[WARNING]:Profiler function <class 'torch.autograd.profiler.record_function'> will be ignored
[DEBUG]:Output code: 

# AOT ID: ['0_inference']
from ctypes import c_void_p, c_long
import torch
import math
import random
import os

mlazos

import torch
import torch._dynamo as torchdynamo
import torch._inductor
import time
import torch._inductor.config as config
from torch._dynamo.utils import cprofile_wrapper
from apex.optimizers import FusedAdam, FusedSGD

config.triton.cudagraphs = True
config.cpp_wrapper = False

kateinoigakukun

Since version 2.23, git-blame has a feature to ignore certain commits. This feature is useful to ignore large formatting or apparently unimportant changes.

How to use

1. Create a revisions list file. The file name is usually .git-blame-ignore-revs
2. Set the file as a default ignore file for blame by git config blame.ignoreRevsFile .git-blame-ignore-revs

The file format is described in git-fsck's man: https://git-scm.com/docs/git-fsck#Documentation/git-fsck.txt-fsckskipList

mcarilli

# This isn't supposed to run as a bash script, i named it with ".sh" for syntax highlighting.

# https://developer.nvidia.com/nsight-systems
# https://docs.nvidia.com/nsight-systems/profiling/index.html

# My preferred nsys (command line executable used to create profiles) commands
#
# In your script, write
# torch.cuda.nvtx.range_push("region name")
# ...

crcrpar

If you want to use VSCode's fantastic feature: Dev Container, put the `devcontainer.json` in `.devcontainer` directory.

imenurok

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import chainer
import chainer.functions as F
import chainer.links as L
import numpy as np

class FRN(chainer.Chain):
    def __init__(self, in_c):

ines

pip install streamlit
pip install spacy
python -m spacy download en_core_web_sm
python -m spacy download en_core_web_md
python -m spacy download de_core_news_sm

thomwolf

def top_k_top_p_filtering(logits, top_k=0, top_p=0.0, filter_value=-float('Inf')):
    """ Filter a distribution of logits using top-k and/or nucleus (top-p) filtering
        Args:
            logits: logits distribution shape (vocabulary size)
            top_k >0: keep only top k tokens with highest probability (top-k filtering).
            top_p >0.0: keep the top tokens with cumulative probability >= top_p (nucleus filtering).
                Nucleus filtering is described in Holtzman et al. (http://arxiv.org/abs/1904.09751)
    """
    assert logits.dim() == 1  # batch size 1 for now - could be updated for more but the code would be less clear
    top_k = min(top_k, logits.size(-1))  # Safety check