Birch-san

Fine-tuning LLaMA-7B on ~12GB VRAM with QLoRA, 4-bit quantization

nvidia-smi said this required 11181MiB, at least to train on the sequence lengths of prompt that occurred initially in the alpaca dataset (~337 token long prompts).
You can get this down to about 10.9GB if (by modifying qlora.py) you run torch.cuda.empty_cache() after PEFT has been applied to your loaded model and before you begin training.

Setup

All instructions are written assuming your command-line shell is bash.

Clone repository:

Dan-Q

#!/bin/bash

# The following steps, which were tested on Ubuntu 18.04 LTS and on the Ubuntu-powered Linux for Windows Subsystem on Windows,
# will:
#
# * Compile a recent version of OpenSSL (you can skip this step and use your package maintainer's version if you prefer, but you
#                                        might have to tweak a few bits)
# * Create a separate set of configuration files suitable for configuring a basic CA capable of signing EV certificates
# * Create such a CA (hackerca.local / HackerCA EV Root CA)
# * Create a certificate request for a site, hackersite.local, belonging to company "Barclays PLC [GB]"

kameshsampath

param(
    [string] $IstioVersion = "0.3.0"
)

$url = "https://github.com/istio/istio/releases/download/$($IstioVersion)/istio_$($IstioVersion)_win.zip"
$Path = Get-Location
$output = [IO.Path]::Combine($Path, "istio_$($IstioVersion)_win.zip”)
    
Write-Host "Downloading Istio from $url to path " $Path -ForegroundColor Green 
    

rveitch

sass/
|
|– base/
|   |– _reset.scss       # Reset/normalize
|   |– _typography.scss  # Typography rules
|   ...                  # Etc…
|
|– components/
|   |– _buttons.scss     # Buttons
|   |– _carousel.scss    # Carousel