lucataco

# Setup:
# conda create -n wan python=3.10
# conda activate wan
# pip3 install torch torchvision torchaudio
# pip install git+https://github.com/huggingface/diffusers.git@3ee899fa0c0a443db371848a87582b2e2295852d 
# pip install accelerate==1.4.0
# pip install transformers==4.49.0
# pip install ftfy==6.3.1

lucataco

services:
  pihole-unbound:
    image: 'bigbeartechworld/big-bear-pihole-unbound:2024.07.0'
    environment:
      - SERVICE_FQDN_PIHOLE_8080
      - SERVICE_FQDN_PIHOLE_10443
      - 'DNS1=127.0.0.1#5353'
      - DNS2=no
      - TZ=America/Chicago
      - WEBPASSWORD=$SERVICE_PASSWORD_PIHOLE

lucataco

from optimum.quanto import freeze, qfloat8, quantize
from diffusers import FluxPipeline
import torch
import time

seed=1337
generator = torch.Generator("cuda").manual_seed(seed)

pipeline = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16).to("cuda")

lucataco

# conda create -n flux python=3.11
# conda activate flux
# pip install torch==2.3.1
# pip install diffusers==0.30.0 transformers==4.43.3
# pip install sentencepiece==0.2.0 accelerate==0.33.0 protobuf==5.27.3

import torch
from diffusers import  FluxPipeline
import diffusers

lucataco

import torch
import random

random.seed(42)
torch.manual_seed(42)


from transformers import LlamaTokenizer, LlamaForCausalLM
model_path = 'openlm-research/open_llama_3b_v2'
tokenizer = LlamaTokenizer.from_pretrained(model_path, legacy=True);

lucataco

build:
  gpu: true
  python_version: "3.10"
  run:
    - apt update -y && apt install -y software-properties-common python3-launchpadlib && apt update -y && add-apt-repository -y ppa:git-core/ppa && apt update -y
    - apt install -y gcc g++ aria2 git git-lfs wget libgl1 libglib2.0-0 ffmpeg cmake libgtk2.0-0 libopenmpi-dev unzip libssl-dev pkg-config tmux ca-certificates
    - wget https://developer.download.nvidia.com/compute/cuda/12.1.0/local_installers/cuda_12.1.0_530.30.02_linux.run && sh cuda_12.1.0_530.30.02_linux.run --silent --toolkit
    - git clone https://github.com/aristocratos/btop /content/btop && cd /content/btop && make && make install PREFIX=/usr && rm -rf /content/btop
    - pip install notebook pyngrok pickleshare ipywidgets
    - pip install https://github.com/camenduru/wheels/releases/download/replicate/vllm-0.4.0.post1-cp310-cp310-linux_x86_64.whl

lucataco

import os
import cv2
import argparse
import numpy as np

MONITOR_W = 38.5


def write_depth(depth, bits=1, reverse=True):
    depth_min = depth.min()

lucataco

""" To use: install Ollama, clone OpenVoice, run this script in the OpenVoice directory
    brew install portaudio
    brew install git-lfs
    git lfs install

    git clone https://github.com/myshell-ai/OpenVoice
    cd OpenVoice
    git clone https://huggingface.co/myshell-ai/OpenVoice
    cp -r OpenVoice/* .
    

lucataco

# from cog import BasePredictor, Input, Path
import os
import cv2
import time
import math
import torch
import numpy as np
from PIL import Image
from glob import glob
from typing import Optional

lucataco

**Goal**: Run benchmarks of SDXL, SVD, and Llama 13B on an L40S test node

**TL;DR**:

- L40S has same inference speeds as A40 for SDXL
- L40S has 10% faster inference speeds than A40S for llama2
- L40S are ~9% faster at Video rendering than A40s

**Process**: Run non-docker/cog python code for fp16