text-generation-webui/Colab-TextGen-GPU.ipynb

118 lines
4.0 KiB
Plaintext
Raw Normal View History

2023-10-22 05:27:52 +02:00
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"private_outputs": true,
"provenance": [],
"gpuType": "T4"
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
},
"accelerator": "GPU"
},
"cells": [
{
"cell_type": "markdown",
"source": [
"# oobabooga/text-generation-webui\n",
"\n",
"After running both cells, a public gradio URL will appear at the bottom in around 10 minutes. You can optionally generate an API link.\n",
2023-10-22 05:27:52 +02:00
"\n",
"* Project page: https://github.com/oobabooga/text-generation-webui\n",
"* Gradio server status: https://status.gradio.app/"
],
"metadata": {
"id": "MFQl6-FjSYtY"
}
},
{
"cell_type": "code",
"source": [
"#@title 1. Keep this tab alive to prevent Colab from disconnecting you { display-mode: \"form\" }\n",
"\n",
"#@markdown Press play on the music player that will appear below:\n",
"%%html\n",
"<audio src=\"https://oobabooga.github.io/silence.m4a\" controls>"
],
"metadata": {
"id": "f7TVVj_z4flw"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"#@title 2. Launch the web UI\n",
"\n",
"#@markdown If unsure about the branch, write \"main\" or leave it blank.\n",
"\n",
"import os\n",
2023-10-22 05:27:52 +02:00
"from pathlib import Path\n",
"\n",
"os.environ.pop('PYTHONPATH', None)\n",
"\n",
2023-10-22 05:27:52 +02:00
"if Path.cwd().name != 'text-generation-webui':\n",
" print(\"\\033[1;32;1m\\n --> Installing the web UI. This will take a while, but after the initial setup, you can download and test as many models as you like.\\033[0;37;0m\\n\")\n",
2023-10-22 05:27:52 +02:00
"\n",
" !git clone https://github.com/oobabooga/text-generation-webui\n",
" %cd text-generation-webui\n",
"\n",
" # Install the project in an isolated environment\n",
" !GPU_CHOICE=A \\\n",
" USE_CUDA118=FALSE \\\n",
" LAUNCH_AFTER_INSTALL=FALSE \\\n",
" INSTALL_EXTENSIONS=FALSE \\\n",
" ./start_linux.sh\n",
2023-10-22 05:27:52 +02:00
"\n",
"# Parameters\n",
"model_url = \"https://huggingface.co/turboderp/gemma-2-9b-it-exl2\" #@param {type:\"string\"}\n",
"branch = \"8.0bpw\" #@param {type:\"string\"}\n",
"command_line_flags = \"--n-gpu-layers 128 --load-in-4bit --use_double_quant --no_flash_attn\" #@param {type:\"string\"}\n",
2023-10-22 05:27:52 +02:00
"api = False #@param {type:\"boolean\"}\n",
"\n",
"if api:\n",
" for param in ['--api', '--public-api']:\n",
" if param not in command_line_flags:\n",
" command_line_flags += f\" {param}\"\n",
"\n",
"model_url = model_url.strip()\n",
2023-10-22 17:57:16 +02:00
"if model_url != \"\":\n",
" if not model_url.startswith('http'):\n",
" model_url = 'https://huggingface.co/' + model_url\n",
"\n",
" # Download the model\n",
" url_parts = model_url.strip('/').strip().split('/')\n",
" output_folder = f\"{url_parts[-2]}_{url_parts[-1]}\"\n",
" branch = branch.strip('\"\\' ')\n",
2023-11-15 17:18:32 +01:00
" if branch.strip() not in ['', 'main']:\n",
2023-10-22 17:57:16 +02:00
" output_folder += f\"_{branch}\"\n",
" !python download-model.py {model_url} --branch {branch}\n",
" else:\n",
" !python download-model.py {model_url}\n",
2023-10-22 05:27:52 +02:00
"else:\n",
2023-10-22 17:57:16 +02:00
" output_folder = \"\"\n",
2023-10-22 05:27:52 +02:00
"\n",
"# Start the web UI\n",
"cmd = f\"./start_linux.sh {command_line_flags} --share\"\n",
2023-10-22 17:57:16 +02:00
"if output_folder != \"\":\n",
" cmd += f\" --model {output_folder}\"\n",
"\n",
2023-10-22 05:27:52 +02:00
"!$cmd"
],
"metadata": {
"id": "LGQ8BiMuXMDG",
"cellView": "form"
},
"execution_count": null,
"outputs": []
}
]
}