File size: 13,403 Bytes
6f8c3bd |
1 |
{"cells": [{"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["\n", "# Welcome to StableTuner, Let's get started!\n", "#### This notebook will guide you through the setup process.\n", "\n", "\n", "__[Join the ST Discord for support, chat and fun times :)](https://discord.gg/DahNECrBUZ)__"]}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["#### Start by uploading your payload.zip file (just drag and drop it to the file area) and run this cell as it gets uploaded."]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": ["#Much thanks to IndustriaDitat and entmike for helping making ST linux compatible!\n", "from IPython.display import clear_output\n", "from subprocess import getoutput\n", "installed_xformers = False\n", "GPU_CardName = getoutput('nvidia-smi --query-gpu=name --format=csv,noheader')\n", "\n", "%pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 --extra-index-url \"https://download.pytorch.org/whl/cu116\"\n", "%pip install -U --pre triton\n", "%pip install ninja bitsandbytes\n", "if '4090' in GPU_CardName:\n", " %pip install https://huggingface.co/industriaditat/xformers_precompiles/resolve/main/RTX4090-xf14-cu116-py38/xformers-0.0.14.dev0-cp38-cp38-linux_x86_64.whl\n", " installed_xformers = True\n", "if '3090' in GPU_CardName:\n", " %pip install https://huggingface.co/industriaditat/xformers_precompiles/resolve/main/RTX3090-xf14-cu116-py38/xformers-0.0.14.dev0-cp38-cp38-linux_x86_64.whl\n", " installed_xformers = True\n", "if 'A5000' in GPU_CardName:\n", " %pip install https://huggingface.co/industriaditat/xformers_precompiles/resolve/main/A5000-xf14-cu116-py38/xformers-0.0.14.dev0-cp38-cp38-linux_x86_64.whl\n", " installed_xformers = True\n", "if 'T4' in GPU_CardName:\n", " %pip install https://huggingface.co/r4ziel/xformers_pre_built/resolve/main/xformers-0.0.14.dev0-cp38-cp38-linux_x86_64_t4.whl\n", " installed_xformers = True\n", "if 'A100' in GPU_CardName:\n", " %pip install https://huggingface.co/industriaditat/xformers_precompiles/resolve/main/A100_13dev/xformers-0.0.13.dev0-py3-none-any.whl\n", " installed_xformers = True\n", "if 'V100' in GPU_CardName:\n", " %pip install https://huggingface.co/industriaditat/xformers_precompiles/resolve/main/V100_13dev/xformers-0.0.13.dev0-py3-none-any.whl\n", " installed_xformers = True\n", "if installed_xformers == False:\n", " clear_output()\n", " print(\"No precompiled xformers found for your GPU. Please wait while we compile xformers for your GPU, this might take 20-40 minutes.\")\n", " %pip install git+https://github.com/facebookresearch/xformers@1d31a3a#egg=xformers\n", "%pip install git+https://github.com/huggingface/diffusers.git@0ca1724#egg=diffusers --force-reinstall\n", "clear_output()\n", "print(\"Done!\")"]}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["#### Upload finished?, time to run this next cell!"]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": ["!unzip -o -q payload.zip\n", "%pip install -r requirements.txt\n", "clear_output()\n", "print(\"Done!\")"]}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["#### Looks like you're done installing, let's get training!\n", "\n"]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": "!accelerate launch --mixed_precision=\"fp16\" scripts/trainer.py \\\n --shuffle_per_epoch \\\n --attention=\"xformers\" \\\n --model_variant=\"base\" \\\n --disable_cudnn_benchmark \\\n --sample_step_interval=500 \\\n --stop_text_encoder_training=15 \\\n --pretrained_model_name_or_path=\"edicamargo/fine_tw2\" \\\n --pretrained_vae_name_or_path=\"\" \\\n --output_dir=\"output/tw3_magic_new\" \\\n --seed=3434555 \\\n --resolution=512 \\\n --train_batch_size=40 \\\n --num_train_epochs=30 \\\n --mixed_precision=\"fp16\" \\\n --use_bucketing \\\n --aspect_mode=\"dynamic\" \\\n --aspect_mode_action_preference=\"add\" \\\n --use_8bit_adam \\\n --gradient_checkpointing \\\n --gradient_accumulation_steps=1 \\\n --learning_rate=2e-6 \\\n --lr_warmup_steps=0 \\\n --lr_scheduler=\"constant\" \\\n --train_text_encoder \\\n --use_image_names_as_captions \\\n --concepts_list=\"stabletune_concept_list.json\" \\\n --num_class_images=200 \\\n --save_every_n_epoch=10 \\\n --n_save_sample=1 \\\n --sample_height=512 \\\n --sample_width=512 \\\n --dataset_repeats=1 \\\n --clip_penultimate"}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["## Model Playground\n", "\n", "#### This is where you can test your model and package it up."]}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["Run this cell and select your output model, you can upload the model to HuggingFace or run the next cell to use the Web UI and play around with your model."]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": ["from ipywidgets import widgets\n", "import os\n", "import glob\n", "from IPython.display import clear_output\n", "import torch\n", "from torch import autocast\n", "from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler\n", "from IPython.display import display\n", "import random\n", "import gradio as gr\n", "from requests import HTTPError\n", "from huggingface_hub import create_repo\n", "from huggingface_hub import HfApi\n", "from huggingface_hub import login\n", "from huggingface_hub import logging\n", "from huggingface_hub.utils import hf_raise_for_status, HfHubHTTPError\n", "def upload_hf(token,repo_name,model_path):\n", " if token == '':\n", " print(\"Please enter your HuggingFace token.\")\n", " return\n", " if repo_name == '':\n", " print(\"Please enter your model name.\")\n", " return\n", " if model_path == '':\n", " print(\"Couldn't find a valid model\")\n", " return\n", " api = login(token=token)\n", " api = HfApi()\n", " hf_username = HfApi().whoami(token)['name']\n", " project_repo = repo_name\n", " try:\n", " create_repo(f'{hf_username}/{project_repo}', repo_type='model',token=token)\n", " except HTTPError as http_e:\n", " if http_e.response.status_code == 409:\n", " print('The repo already exists')\n", " pass\n", " else:\n", " print(f'An error occurred: {http_e.reason}')\n", " except HfHubHTTPError as hfhub_e:\n", " if hfhub_e.response.status_code == 409:\n", " pass\n", " else:\n", " print(f'An error occurred: {hfhub_e.message}')\n", " try:\n", " print('Uploading...')\n", " api.upload_folder(\n", " folder_path=model_path,\n", " path_in_repo='',\n", " repo_id=f'{hf_username}/{project_repo}',\n", " repo_type=\"model\",\n", " ignore_patterns=\"**/logs/\",\n", " )\n", " print('Done!')\n", " print(f'Model is at https://huggingface.co/{hf_username}/{project_repo}')\n", " except Exception as general_e:\n", " print(f'Exception occurred: {general_e}')\n", "if 'output' not in os.listdir():\n", " print(\"No output folder found. Please run the training cell first.\")\n", "models = []\n", "model_dir = os.listdir('output')[0]\n", "output_sort = sorted(glob.iglob('output' + os.sep + model_dir + os.sep+ '*'), key=os.path.getctime, reverse=True)\n", "if len(output_sort) == 0:\n", " print(\"No models found in output folder. Please run the training cell first.\")\n", "for model in output_sort:\n", " required_folders = [\"vae\", \"unet\", \"tokenizer\", \"text_encoder\"]\n", " if all(x in os.listdir(model) for x in required_folders):\n", " models.append(model)\n", "model_selection = widgets.Dropdown(\n", " layout={'width': 'initial'},\n", " style={'description_width': 'initial'},\n", " options=models,\n", " value=models[0],\n", " # rows=10,\n", " description='Select Checkpoint:',\n", " disabled=False\n", ")\n", "upload_btn = widgets.Button(\n", " description='Upload to HuggingFace Hub',\n", " style={'description_width': 'initial'},\n", " layout={'width': 'initial'},\n", " disabled=False,\n", " button_style='', # 'success', 'info', 'warning', 'danger' or ''\n", " tooltip='Press to start upload',\n", " icon='check' # (FontAwesome names without the `fa-` prefix)\n", ")\n", "token_txt = widgets.Text(\n", " value='',\n", " style={'description_width': 'initial'},\n", " layout={'width': 'initial'},\n", " placeholder='HF Token',\n", " description='Hugging Face Token:',\n", " disabled=False\n", ")\n", "repo_txt = widgets.Text(\n", " value=model_dir,\n", " style={'description_width': 'initial'},\n", " placeholder='Give your model a name',\n", " description='Model Name:',\n", " disabled=False\n", ")\n", "upload_btn.on_click(lambda x: upload_hf(token_txt.value,repo_txt.value,model_selection.value))\n", "clear_output()\n", "display(model_selection)\n", "display(token_txt)\n", "display(repo_txt)\n", "display(upload_btn)\n", "print('You can input your HuggingFace token and repo to upload your model to the HuggingFace Hub, make sure to use a write API token!')\n", "print('Alternatively, you can run the next cell to open up a UI where you can generate and zip up the model.')\n"]}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["All ready?, lets play around with it ;), this next cell will load a small UI for you to generate images and zip it up for download, re-run this cell if you selected a new model!"]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": ["model_path = model_selection.value\n", "pipe = StableDiffusionPipeline.from_pretrained(model_path,safety_checker=None, torch_dtype=torch.float16).to(\"cuda\")\n", "scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)\n", "pipe.scheduler = scheduler\n", "print('Loaded checkpoint')\n", "def inference(prompt, negative_prompt, num_samples, height=512, width=512, num_inference_steps=50,seed=-1,guidance_scale=7.5):\n", " with torch.autocast(\"cuda\"), torch.inference_mode():\n", " if seed != -1:\n", " g_cuda = torch.Generator(device='cuda')\n", " g_cuda.manual_seed(int(seed))\n", " else:\n", " seed = random.randint(0, 100000)\n", " g_cuda = torch.Generator(device='cuda')\n", " g_cuda.manual_seed(seed)\n", " return pipe(\n", " prompt, height=int(height), width=int(width),\n", " negative_prompt=negative_prompt,\n", " num_images_per_prompt=int(num_samples),\n", " num_inference_steps=int(num_inference_steps), guidance_scale=guidance_scale,\n", " generator=g_cuda\n", " ).images, seed\n", "def zip_model():\n", " import shutil\n", " print('Zipping Model!, Please wait until you see a done message, this can take a few minutes, you can keep generating while you wait!')\n", " curLocation = os.getcwd()\n", " model_name = os.path.dirname(model_path)\n", " shutil.make_archive(model_name,'zip',model_path)\n", " os.chdir(curLocation)\n", " print('Done!')\n", "with gr.Blocks() as demo:\n", " with gr.Row():\n", " with gr.Column():\n", " prompt = gr.Textbox(label=\"Prompt\", value=\"photo of zwx dog in a bucket\")\n", " negative_prompt = gr.Textbox(label=\"Negative Prompt\", value=\"\")\n", " with gr.Row():\n", " run = gr.Button(value=\"Generate\")\n", " zip = gr.Button(value=\"Zip Model For Download\")\n", " with gr.Row():\n", " num_samples = gr.Number(label=\"Number of Samples\", value=4)\n", " guidance_scale = gr.Number(label=\"Guidance Scale\", value=7.5)\n", " with gr.Row():\n", " height = gr.Number(label=\"Height\", value=512)\n", " width = gr.Number(label=\"Width\", value=512)\n", " with gr.Row():\n", " num_inference_steps = gr.Slider(label=\"Steps\", value=25)\n", " seed = gr.Number(label=\"Seed\", value=-1)\n", " with gr.Column():\n", " gallery = gr.Gallery()\n", " seedDisplay = gr.Number(label=\"Used Seed:\", value=0)\n", "\n", " run.click(inference, inputs=[prompt, negative_prompt, num_samples, height, width, num_inference_steps,seed, guidance_scale], outputs=[gallery,seedDisplay])\n", " zip.click(zip_model)\n", "demo.launch(debug=True,share=True)"]}, {"attachments": {}, "cell_type": "markdown", "metadata": {}, "source": ["### This is the end, for now :) ,you can convert your model to CKPT back in StableTuner!."]}], "metadata": {"kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"}, "language_info": {"codemirror_mode": {"name": "ipython", "version": 3}, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.9 (tags/v3.10.9:1dd9be6, Dec 6 2022, 20:01:21) [MSC v.1934 64 bit (AMD64)]"}, "orig_nbformat": 4, "vscode": {"interpreter": {"hash": "886cb931ea414ad2a87adcccbb1ce9166879eb6056301acd331591c6290ceca8"}}}, "nbformat": 4, "nbformat_minor": 2} |