1
0
Fork 0
ai-doc/src/LLM.tex

215 lines
7.8 KiB
TeX
Raw Normal View History

2023-07-19 11:48:36 -06:00
%
% LLM.tex
%
% AI Documentation
%
% Copyright (C) 2022, 2023, Jeff Moe
%
% This document is licensed under the Creative Commons Attribution 4.0
% International Public License (CC BY-SA 4.0) by Jeff Moe.
%
\section{Introduction}
Large Language Model, aka LLM.
\begin{mdframed}[backgroundcolor=blue!10,linecolor=blue!30]
\begin{itemize}
\item BLOOM --- \url{https://huggingface.co/bigscience/bloom}
\url{https://bigscience.huggingface.co/blog/bloom}
\item PolyGlot --- \url{https://github.com/EleutherAI/polyglot}
\item Maxtext --- \url{https://github.com/EleutherAI/maxtext}
\item H2O.ai --- \url{https://h2o.ai/}
\end{itemize}
\end{mdframed}
2023-10-17 13:07:03 -06:00
2023-10-17 13:11:06 -06:00
\section{LoLLMS}
Web gui for LLM.
\begin{minted}{sh}
mkdir -p ~/devel/ParisNeo
cd ~/devel/ParisNeo
git clone --recursive https://github.com/ParisNeo/lollms-webui
cd lollms-webui/
pyenv local 3.11.6
# it likes name env for environment
virtualenv -p 3.11.6 env
source env/bin/activate
pip install -U setuptools wheel pip
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
~/devel/pytorch/verify.py
pip install -r requirements.txt
pip install exllamav2
bash webui.sh
# * When script prompts, put dir here:
# `~/devel/ParisNeo/lollms`
#
# * In `~/devel/ParisNeo/lollms/config` set IP to 0.0.0.0.
#
# * Open firewall.
\end{minted}
2023-10-17 13:07:03 -06:00
\section{Llama}
Llama very nice creature. Is also LLM.
\begin{minted}{sh}
# misc
apt update
apt -y install git git-lfs cmake vim-tiny cmake-curses-gui libncurses-dev
# nvtop
git clone https://github.com/Syllo/nvtop
cd nvtop
mkdir build
cd build
cmake ..
ccmake ..
make -j`nproc`
./src/nvtop
# go
wget https://go.dev/dl/go1.21.2.linux-amd64.tar.gz
#sudo rm -rf /usr/local/go && sudo tar -C /usr/local -xzf go1.21.2.linux-amd64.tar.gz
rm -rf /usr/local/go && tar -C /usr/local -xzf go1.21.2.linux-amd64.tar.gz
echo 'export PATH=/usr/local/go/bin:$PATH' >> ~/.bashrc
# Build
# Maybe for ppc64le, but doesn't work:
export CMAKE_C_COMPILER=/usr/lib/ccache/gcc-11
export CMAKE_CXX_COMPILER=/usr/lib/ccache/g++-11
export CUDAToolkit_INCLUDE_DIR=/usr/include
# Total crap that does work (on morvolta):
cd /usr/lib
ln -s /usr/include .
git clone --recursive https://github.com/jmorganca/ollama
cd ollama/
go clean
go generate ./...
go build .
# Run
# set up podrun dirs:
mkdir /workspace/ollama
ln -s /workspace/ollama .ollama
./olama server
./ollama pull wizardcoder:34b-python
# ollama pull wizardcoder:7b-python
# ollama pull wizardcoder:13b-python
# ollama pull wizardcoder:34b-python
# ollama pull wizardlm:70b-llama2-q4_0
./ollama run wizardcoder:34b-python
ollama pull wizardcoder:34b-python
ollama pull wizardcoder:34b-python-q2_K
ollama pull wizardcoder:34b-python-q3_K_L
ollama pull wizardcoder:34b-python-q3_K_M
ollama pull wizardcoder:34b-python-q3_K_S
ollama pull wizardcoder:34b-python-q4_0
ollama pull wizardcoder:34b-python-q4_1
ollama pull wizardcoder:34b-python-q4_K_M
ollama pull wizardcoder:34b-python-q4_K_S
ollama pull wizardcoder:34b-python-q5_0
ollama pull wizardcoder:34b-python-q5_1
ollama pull wizardcoder:34b-python-q5_K_M
ollama pull wizardcoder:34b-python-q5_K_S
ollama pull wizardcoder:34b-python-q6_K
ollama pull wizardcoder:34b-python-q8_0
ollama run wizardcoder:34b-python
ollama run wizardcoder:34b-python-q2_K
ollama run wizardcoder:34b-python-q3_K_L
ollama run wizardcoder:34b-python-q3_K_M
ollama run wizardcoder:34b-python-q3_K_S
ollama run wizardcoder:34b-python-q4_0
ollama run wizardcoder:34b-python-q4_1
ollama run wizardcoder:34b-python-q4_K_M
ollama run wizardcoder:34b-python-q4_K_S
ollama run wizardcoder:34b-python-q5_0
ollama run wizardcoder:34b-python-q5_1
ollama run wizardcoder:34b-python-q5_K_M
ollama run wizardcoder:34b-python-q5_K_S
ollama run wizardcoder:34b-python-q6_K
ollama run wizardcoder:34b-python-q8_0
ollama pull orca-mini
$ ollama list
NAME ID SIZE MODIFIED
codellama:34b-code d78387764871 19 GB 3 days ago
codellama:34b-code-q8_0 2b6690e05081 36 GB 4 days ago
codellama:34b-instruct-q8_0 bf3f9995e93b 36 GB 3 days ago
codellama:34b-python 5ede7f0ac6c4 19 GB 3 days ago
codellama:34b-python-q8_0 73563ada07a6 36 GB 4 days ago
everythinglm:13b-16k-q8_0 9ef6e6d7446f 14 GB 4 days ago
falcon:180b 90d369418a4f 102 GB 3 days ago
falcon:180b-chat-q4_0 90d369418a4f 102 GB 3 days ago
falcon:180b-text-q4_0 aca441c3e642 102 GB 3 days ago
falcon:180b-text-q8_0 ad08feb8304f 191 GB 3 days ago
falcon:40b 2d9a4bfc8555 24 GB 3 days ago
falcon:40b-instruct 2d9a4bfc8555 24 GB 3 days ago
falcon:40b-instruct-fp16 7cbd92dfea70 84 GB 3 days ago
falcon:40b-text 77ecf2f4218a 24 GB 3 days ago
falcon:40b-text-fp16 c42691ec8c94 84 GB 3 days ago
llama2:70b f60ae38a353b 39 GB 3 days ago
mistral:7b-instruct-q8_0 f97a185cacf4 7.7 GB 4 days ago
mistral:7b-text-q8_0 e99f29355cda 7.7 GB 4 days ago
nexusraven:13b 336957c1d527 7.4 GB 3 days ago
nexusraven:13b-q8_0 25fba36ef0af 14 GB 3 days ago
orca-mini:13b-v3 5bc199b18569 7.4 GB 43 hours ago
orca-mini:70b-v3 179d05395377 39 GB 43 hours ago
orca-mini:7b-v3 de4ca17ad6a7 3.8 GB 44 hours ago
orca-mini:latest 5e9bc249c869 1.9 GB 4 days ago
phind-codellama:34b-python 4f719aca701f 19 GB 3 days ago
phind-codellama:34b-python-q8_0 9e8121c0c614 36 GB 3 days ago
phind-codellama:34b-v2 e2b45b77c8bf 19 GB 4 days ago
phind-codellama:34b-v2-q2_K fa4f2f18144d 14 GB 3 days ago
phind-codellama:34b-v2-q4_0 e2b45b77c8bf 19 GB 3 days ago
phind-codellama:34b-v2-q4_1 8870ba145794 21 GB 3 days ago
phind-codellama:34b-v2-q4_K_S 9c3bbb7e9ad4 19 GB 3 days ago
phind-codellama:34b-v2-q6_K b20c5fb7a66d 28 GB 3 days ago
phind-codellama:34b-v2-q8_0 1f6f3dca7bbc 36 GB 3 days ago
sqlcoder:15b c06a24fb83df 9.0 GB 3 days ago
sqlcoder:15b-fp16 c5e9fd3852f9 32 GB 3 days ago
sqlcoder:15b-q8_0 7cab292fc701 17 GB 3 days ago
starcoder:15b-base-q8_0 dbe41581594a 17 GB 4 days ago
vicuna:33b 1a7295496c4f 18 GB 3 days ago
vicuna:33b-q8_0 3c52edc13a77 35 GB 3 days ago
wizard-math:70b ec65b71e5de1 39 GB 3 days ago
wizard-math:70b-q5_K_M c14e6fd7faea 49 GB 3 days ago
wizardcoder:13b-python 767adb2cd4cc 7.3 GB 5 days ago
wizardcoder:34b-python 7bd03d2d0b8d 19 GB 8 days ago
wizardcoder:34b-python-q2_K 36a635134a0a 14 GB 5 days ago
wizardcoder:34b-python-q6_K 0c1c3994fb50 28 GB 5 days ago
wizardcoder:34b-python-q8_0 86d30a8b017c 36 GB 4 days ago
wizardcoder:7b-python ae6c60a3e5ce 3.8 GB 4 days ago
\end{minted}
2023-10-17 13:23:48 -06:00
\section{Axolotl}
Best beast ever.
\begin{minted}{sh}
mkdir -p ~/devel/OpenAccess-AI-Collective
cd ~/devel/OpenAccess-AI-Collective/
pyenv local 3.10
virtualenv -p 3.10 venv
source venv/bin/activate
pip install -U setuptools pip wheel
pip install torch==2.0.1+cu118 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip3 install packaging
pip3 install -e '.[flash-attn,deepspeed]'
pip3 install -U git+https://github.com/huggingface/peft.git
accelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml
accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml --lora_model_dir="./lora-out"
\end{minted}