jondurbin

Trained on 10x a6000 GPUs on runpod.io.

I actually ran many fine-tunes, including multiple full-finetunes, fp16 loras, and qloras, and the below qlora actually did best in my testing.

dataset: https://hf.co/datasets/jondurbin/airoboros-3.1 (plus a few unpublished de-censoring instructions)

training script: https://github.com/jondurbin/qlora specifically commit 8cd269bf9bd7753c92164934269019e12f23314f

export BASE_DIR=/workspace

ravibhure

In your local clone of your forked repository, you can add the original GitHub repository as a "remote". ("Remotes" are like nicknames for the URLs of repositories - origin is one, for example.) Then you can fetch all the branches from that upstream repository, and rebase your work to continue working on the upstream version. In terms of commands that might look like:

Add the remote, call it "upstream":

git remote add upstream https://github.com/whoever/whatever.git

Fetch all the branches of that remote into remote-tracking branches, such as upstream/master:

git fetch upstream

tokestermw

import json
import urlparse
from itertools import chain
flatten = chain.from_iterable

from nltk import word_tokenize

from gensim.corpora import Dictionary
from gensim.models.ldamodel import LdaModel
from gensim.models.tfidfmodel import TfidfModel

graydon

# extracted from http//www.naturalearthdata.com/download/110m/cultural/ne_110m_admin_0_countries.zip
# under public domain terms

country_bounding_boxes = {
    'AF': ('Afghanistan', (60.5284298033, 29.318572496, 75.1580277851, 38.4862816432)),
    'AO': ('Angola', (11.6400960629, -17.9306364885, 24.0799052263, -4.43802336998)),
    'AL': ('Albania', (19.3044861183, 39.624997667, 21.0200403175, 42.6882473822)),
    'AE': ('United Arab Emirates', (51.5795186705, 22.4969475367, 56.3968473651, 26.055464179)),
    'AR': ('Argentina', (-73.4154357571, -55.25, -53.628348965, -21.8323104794)),
    'AM': ('Armenia', (43.5827458026, 38.7412014837, 46.5057198423, 41.2481285671)),