Good question! I am collecting human data on how quantization affects outputs. See here for more information: ggml-org/llama.cpp#5962
In the meantime, use the largest that fully fits in your GPU. If you can comfortably fit Q4_K_S, try using a model with more parameters.
See the wiki upstream: https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import time | |
from contextlib import suppress | |
import torch | |
import torch.nn as nn | |
import torch.optim as optim | |
import torch.nn.functional as F | |
import torch.backends.cuda as cuda | |
from torch.utils.data import DataLoader, IterableDataset |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Android Emulator (ARM64) on EC2 - 2022 | |
--------------------------------------- | |
1. Launch EC2 ARM based Instance (a1.metal / a1.2xlarge): (16 Gb RAM, 32Gb Disk), Ubuntu Server 22.04 LTS (HVM) ARM x64 | |
2. sudo apt update && sudo apt upgrade | |
3. sudo apt install default-jdk python3-pip repo python-is-python3 unzip libpcre2-dev adb | |
4. wget https://dl.google.com/android/repository/commandlinetools-linux-8512546_latest.zip | |
5. unzip commandlinetools-linux-8512546_latest.zip -d android-sdk | |
6. sudo mv android-sdk /opt/ | |
7. mkdir /opt/android-sdk/cmdline-tools/latest | |
8. mv /opt/android-sdk/cmdline-tools/* /opt/android-sdk/cmdline-tools/latest (ignore the error) |
Steps below iterate how to manually recreate the disk partitions required by Windows and restore a Windows installation from a system image backup created with the "Backup and Restore (Windows 7)" feature, especially when the official steps fail.
These steps were tested with Windows 11, but should work with prior versions of Windows as well:
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# Windows: Create a shortcut to a .bat file in 'shell:sendto' to call this script with files passed as arguments | |
# Use with 'save_proxy_clip_list.py' gist to quickly pull used timeline clips into FFMPEG. | |
# Use 'link_proxies.py' gist to relink proxies correctly. | |
# Bug in current Resolve release links clips wrongly to one or two proxies only. | |
# This assumes FFMPEG is on path. | |
import os, sys, shutil | |
import subprocess | |
import argparse | |
import pathlib |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# YouTube (english) : https://www.youtube.com/watch?v=FtU2_bBfSgM | |
# YouTube (french) : https://www.youtube.com/watch?v=VjnaVBnERDU | |
# | |
# On your laptop, connect to the Mac instance with SSH (similar to Linux instances) | |
# | |
ssh -i <your private key.pem> ec2-user@<your public ip address> | |
# | |
# On the Mac |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
<body onload=z=c.getContext`2d`,setInterval(`c.width=W=150,Y<W&&P<Y&Y<P+E|9<p?z.fillText(S++${Y=`,9,9|z.fillRect(p`}*0,Y-=--M${Y+Y},P+E,9,W),P))):p=M=Y=S=6,p=p-6||(P=S%E,W)`,E=49) onclick=M=9><canvas id=c> |
NewerOlder