{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"import gradio as gr\n",
"import pandas as pd\n"
]
},
{
"cell_type": "code",
"execution_count": 59,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"df = pd.read_csv(\n",
" \"https://docs.google.com/spreadsheets/d/e/2PACX-1vSC40sszorOjHfozmNqJT9lFiJhG94u3fbr3Ss_7fzcU3xqqJQuW1Ie_SNcWEB-uIsBi9NBUK7-ddet/pub?output=csv\",\n",
" skiprows=1,\n",
")\n"
]
},
{
"cell_type": "code",
"execution_count": 60,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"# Drop footers\n",
"df = df.copy()[~df[\"Model\"].isna()]\n"
]
},
{
"cell_type": "code",
"execution_count": 61,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"# Drop TBA models\n",
"df = df.copy()[df[\"Parameters \\n(B)\"] != \"TBA\"]\n"
]
},
{
"cell_type": "code",
"execution_count": 62,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"
\n",
"\n",
"
\n",
" \n",
" \n",
" | \n",
" Model | \n",
" Lab | \n",
" Selected \\nplaygrounds | \n",
" Parameters \\n(B) | \n",
" Tokens \\ntrained (B) | \n",
" Ratio T:P\\n(Chinchilla scaling) | \n",
" Training dataset | \n",
" Announced\\nβΌ | \n",
" Public? | \n",
" Released | \n",
" Paper / Repo | \n",
" Notes | \n",
"
\n",
" \n",
" \n",
" \n",
" 3 | \n",
" Kosmos-1 | \n",
" Microsoft | \n",
" NaN | \n",
" 1.6 | \n",
" 360 | \n",
" 225:1 | \n",
" π πβ¬ πΈ π | \n",
" Feb/2023 | \n",
" π΄ | \n",
" Feb/2023 | \n",
" https://arxiv.org/abs/2302.14045 | \n",
" Multimodal large language model (MLLM). Ravenβ... | \n",
"
\n",
" \n",
" 4 | \n",
" LLaMA-65B | \n",
" Meta AI | \n",
" Weights leaked: https://github.com/facebookres... | \n",
" 65 | \n",
" 1400 | \n",
" 22:1 | \n",
" π πβ¬ πΈ π | \n",
" Feb/2023 | \n",
" π‘ | \n",
" Feb/2023 | \n",
" https://research.facebook.com/publications/lla... | \n",
" Researchers only, noncommercial only. 'LLaMA-6... | \n",
"
\n",
" \n",
" 5 | \n",
" MOSS | \n",
" Fudan University | \n",
" https://moss.fastnlp.top/ | \n",
" 20 | \n",
" 430 | \n",
" 22:1 | \n",
" πΈ π | \n",
" Feb/2023 | \n",
" π’ | \n",
" Feb/2023 | \n",
" https://txsun1997.github.io/blogs/moss.html | \n",
" Major bandwidth issues: https://www.reuters.co... | \n",
"
\n",
" \n",
" 6 | \n",
" Palmyra | \n",
" Writer | \n",
" https://huggingface.co/models?search=palmyra | \n",
" 20 | \n",
" 300 | \n",
" 15:1 | \n",
" π | \n",
" Feb/2023 | \n",
" π’ | \n",
" Feb/2023 | \n",
" https://writer.com/blog/palmyra/ | \n",
" Only up to 5B available open-source 'trained o... | \n",
"
\n",
" \n",
" 7 | \n",
" Luminous Supreme Control | \n",
" Aleph Alpha | \n",
" https://app.aleph-alpha.com/playground/completion | \n",
" 70 | \n",
" NaN | \n",
" NaN | \n",
" π πβ¬ πΈ π₯ | \n",
" Feb/2023 | \n",
" π’ | \n",
" Feb/2023 | \n",
" https://docs.aleph-alpha.com/docs/introduction... | \n",
" βControlβ means instruction tuned | \n",
"
\n",
" \n",
"
\n",
"
"
],
"text/plain": [
" Model Lab \\\n",
"3 Kosmos-1 Microsoft \n",
"4 LLaMA-65B Meta AI \n",
"5 MOSS Fudan University \n",
"6 Palmyra Writer \n",
"7 Luminous Supreme Control Aleph Alpha \n",
"\n",
" Selected \\nplaygrounds Parameters \\n(B) \\\n",
"3 NaN 1.6 \n",
"4 Weights leaked: https://github.com/facebookres... 65 \n",
"5 https://moss.fastnlp.top/ 20 \n",
"6 https://huggingface.co/models?search=palmyra 20 \n",
"7 https://app.aleph-alpha.com/playground/completion 70 \n",
"\n",
" Tokens \\ntrained (B) Ratio T:P\\n(Chinchilla scaling) Training dataset \\\n",
"3 360 225:1 π πβ¬ πΈ π \n",
"4 1400 22:1 π πβ¬ πΈ π \n",
"5 430 22:1 πΈ π \n",
"6 300 15:1 π \n",
"7 NaN NaN π πβ¬ πΈ π₯ \n",
"\n",
" Announced\\nβΌ Public? Released \\\n",
"3 Feb/2023 π΄ Feb/2023 \n",
"4 Feb/2023 π‘ Feb/2023 \n",
"5 Feb/2023 π’ Feb/2023 \n",
"6 Feb/2023 π’ Feb/2023 \n",
"7 Feb/2023 π’ Feb/2023 \n",
"\n",
" Paper / Repo \\\n",
"3 https://arxiv.org/abs/2302.14045 \n",
"4 https://research.facebook.com/publications/lla... \n",
"5 https://txsun1997.github.io/blogs/moss.html \n",
"6 https://writer.com/blog/palmyra/ \n",
"7 https://docs.aleph-alpha.com/docs/introduction... \n",
"\n",
" Notes \n",
"3 Multimodal large language model (MLLM). Ravenβ... \n",
"4 Researchers only, noncommercial only. 'LLaMA-6... \n",
"5 Major bandwidth issues: https://www.reuters.co... \n",
"6 Only up to 5B available open-source 'trained o... \n",
"7 βControlβ means instruction tuned "
]
},
"execution_count": 62,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.head()\n"
]
},
{
"cell_type": "code",
"execution_count": 63,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"\n",
"
\n",
" \n",
" \n",
" | \n",
" Model | \n",
" Lab | \n",
" Selected \\nplaygrounds | \n",
" Parameters \\n(B) | \n",
" Tokens \\ntrained (B) | \n",
" Ratio T:P\\n(Chinchilla scaling) | \n",
" Training dataset | \n",
" Announced\\nβΌ | \n",
" Public? | \n",
" Released | \n",
" Paper / Repo | \n",
" Notes | \n",
"
\n",
" \n",
" \n",
" \n",
" 92 | \n",
" Meena | \n",
" Google | \n",
" NaN | \n",
" 2.6 | \n",
" 10000 | \n",
" 3,847:1 | \n",
" π₯ π | \n",
" Jan/2020 | \n",
" π΄ | \n",
" Jan/2020 | \n",
" https://arxiv.org/abs/2001.09977 | \n",
" Dialogue model. Trained 61B tokens for 164x ep... | \n",
"
\n",
" \n",
" 93 | \n",
" RoBERTa | \n",
" Meta AI | \n",
" Hugging Face | \n",
" 0.355 | \n",
" 2200 | \n",
" 6,198:1 | \n",
" π π β¬ πΈ | \n",
" Jul/2019 | \n",
" π’ | \n",
" Jul/2019 | \n",
" https://arxiv.org/abs/1907.11692 | \n",
" See cite ROBERTA | \n",
"
\n",
" \n",
" 94 | \n",
" GPT-2 | \n",
" OpenAI | \n",
" Hugging Face | \n",
" 1.5 | \n",
" 10 | \n",
" 7:1 | \n",
" β¬ | \n",
" Feb/2019 | \n",
" π’ | \n",
" Nov/2019 | \n",
" https://openai.com/blog/better-language-models/ | \n",
" Reddit outbound only | \n",
"
\n",
" \n",
" 95 | \n",
" GPT-1 | \n",
" OpenAI | \n",
" Hugging Face | \n",
" 0.1 | \n",
" NaN | \n",
" NaN | \n",
" π | \n",
" Jun/2018 | \n",
" π’ | \n",
" Jun/2018 | \n",
" https://openai.com/blog/language-unsupervised/ | \n",
" Books only | \n",
"
\n",
" \n",
" 96 | \n",
" BERT | \n",
" Google | \n",
" Hugging Face | \n",
" 0.3 | \n",
" 137 | \n",
" 457:1 | \n",
" π π | \n",
" Oct/2018 | \n",
" π’ | \n",
" Oct/2018 | \n",
" https://arxiv.org/abs/1810.04805 | \n",
" NaN | \n",
"
\n",
" \n",
"
\n",
"
"
],
"text/plain": [
" Model Lab Selected \\nplaygrounds Parameters \\n(B) \\\n",
"92 Meena Google NaN 2.6 \n",
"93 RoBERTa Meta AI Hugging Face 0.355 \n",
"94 GPT-2 OpenAI Hugging Face 1.5 \n",
"95 GPT-1 OpenAI Hugging Face 0.1 \n",
"96 BERT Google Hugging Face 0.3 \n",
"\n",
" Tokens \\ntrained (B) Ratio T:P\\n(Chinchilla scaling) Training dataset \\\n",
"92 10000 3,847:1 π₯ π \n",
"93 2200 6,198:1 π π β¬ πΈ \n",
"94 10 7:1 β¬ \n",
"95 NaN NaN π \n",
"96 137 457:1 π π \n",
"\n",
" Announced\\nβΌ Public? Released \\\n",
"92 Jan/2020 π΄ Jan/2020 \n",
"93 Jul/2019 π’ Jul/2019 \n",
"94 Feb/2019 π’ Nov/2019 \n",
"95 Jun/2018 π’ Jun/2018 \n",
"96 Oct/2018 π’ Oct/2018 \n",
"\n",
" Paper / Repo \\\n",
"92 https://arxiv.org/abs/2001.09977 \n",
"93 https://arxiv.org/abs/1907.11692 \n",
"94 https://openai.com/blog/better-language-models/ \n",
"95 https://openai.com/blog/language-unsupervised/ \n",
"96 https://arxiv.org/abs/1810.04805 \n",
"\n",
" Notes \n",
"92 Dialogue model. Trained 61B tokens for 164x ep... \n",
"93 See cite ROBERTA \n",
"94 Reddit outbound only \n",
"95 Books only \n",
"96 NaN "
]
},
"execution_count": 63,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.tail()\n"
]
},
{
"cell_type": "code",
"execution_count": 64,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"def make_clickable_cell(cell):\n",
" if pd.isnull(cell):\n",
" return \"\"\n",
" else:\n",
" return f'{cell}'\n"
]
},
{
"cell_type": "code",
"execution_count": 65,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"columns_to_click = [\"Paper / Repo\", \"Selected \\nplaygrounds\"]\n",
"for col in columns_to_click:\n",
" df[col] = df[col].apply(make_clickable_cell)\n"
]
},
{
"cell_type": "code",
"execution_count": 66,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"\n",
"
\n",
" \n",
" \n",
" | \n",
" Model | \n",
" Lab | \n",
" Selected \\nplaygrounds | \n",
" Parameters \\n(B) | \n",
" Tokens \\ntrained (B) | \n",
" Ratio T:P\\n(Chinchilla scaling) | \n",
" Training dataset | \n",
" Announced\\nβΌ | \n",
" Public? | \n",
" Released | \n",
" Paper / Repo | \n",
" Notes | \n",
"
\n",
" \n",
" \n",
" \n",
" 3 | \n",
" Kosmos-1 | \n",
" Microsoft | \n",
" | \n",
" 1.6 | \n",
" 360 | \n",
" 225:1 | \n",
" π πβ¬ πΈ π | \n",
" Feb/2023 | \n",
" π΄ | \n",
" Feb/2023 | \n",
" <a target=\"_blank\" href=\"https://arxiv.org/ab... | \n",
" Multimodal large language model (MLLM). Ravenβ... | \n",
"
\n",
" \n",
" 4 | \n",
" LLaMA-65B | \n",
" Meta AI | \n",
" <a target=\"_blank\" href=\"Weights leaked: http... | \n",
" 65 | \n",
" 1400 | \n",
" 22:1 | \n",
" π πβ¬ πΈ π | \n",
" Feb/2023 | \n",
" π‘ | \n",
" Feb/2023 | \n",
" <a target=\"_blank\" href=\"https://research.fac... | \n",
" Researchers only, noncommercial only. 'LLaMA-6... | \n",
"
\n",
" \n",
" 5 | \n",
" MOSS | \n",
" Fudan University | \n",
" <a target=\"_blank\" href=\"https://moss.fastnlp... | \n",
" 20 | \n",
" 430 | \n",
" 22:1 | \n",
" πΈ π | \n",
" Feb/2023 | \n",
" π’ | \n",
" Feb/2023 | \n",
" <a target=\"_blank\" href=\"https://txsun1997.gi... | \n",
" Major bandwidth issues: https://www.reuters.co... | \n",
"
\n",
" \n",
" 6 | \n",
" Palmyra | \n",
" Writer | \n",
" <a target=\"_blank\" href=\"https://huggingface.... | \n",
" 20 | \n",
" 300 | \n",
" 15:1 | \n",
" π | \n",
" Feb/2023 | \n",
" π’ | \n",
" Feb/2023 | \n",
" <a target=\"_blank\" href=\"https://writer.com/b... | \n",
" Only up to 5B available open-source 'trained o... | \n",
"
\n",
" \n",
" 7 | \n",
" Luminous Supreme Control | \n",
" Aleph Alpha | \n",
" <a target=\"_blank\" href=\"https://app.aleph-al... | \n",
" 70 | \n",
" NaN | \n",
" NaN | \n",
" π πβ¬ πΈ π₯ | \n",
" Feb/2023 | \n",
" π’ | \n",
" Feb/2023 | \n",
" <a target=\"_blank\" href=\"https://docs.aleph-a... | \n",
" βControlβ means instruction tuned | \n",
"
\n",
" \n",
"
\n",
"
"
],
"text/plain": [
" Model Lab \\\n",
"3 Kosmos-1 Microsoft \n",
"4 LLaMA-65B Meta AI \n",
"5 MOSS Fudan University \n",
"6 Palmyra Writer \n",
"7 Luminous Supreme Control Aleph Alpha \n",
"\n",
" Selected \\nplaygrounds Parameters \\n(B) \\\n",
"3 1.6 \n",
"4 The Large Language Models Landscape\"\"\"\n",
"description = \"\"\"Large Language Models (LLMs) today come in a variety architectures and capabilities. This interactive landscape provides a visual overview of the most important LLMs, including their training data, size, release date, and whether they are openly accessible or not. It also includes notes on each model to provide additional context. This landscape is derived from data compiled by Dr. Alan D. Thompson at [lifearchitect.ai](https://lifearchitect.ai).\n",
"\"\"\"\n"
]
},
{
"cell_type": "code",
"execution_count": 68,
"metadata": {},
"outputs": [],
"source": [
"# |export\n",
"dtypes = [\"str\" if c not in columns_to_click else \"markdown\" for c in df.columns]\n"
]
},
{
"cell_type": "code",
"execution_count": 69,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Running on local URL: http://127.0.0.1:7868\n",
"\n",
"To create a public link, set `share=True` in `launch()`.\n"
]
},
{
"data": {
"text/html": [
""
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/plain": []
},
"execution_count": 69,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# |export\n",
"def value_func():\n",
" return df\n",
"\n",
"\n",
"with gr.Blocks() as demo:\n",
" gr.Markdown(title)\n",
" gr.Markdown(description)\n",
" gr.components.DataFrame(value=value_func, datatype=dtypes)\n",
"\n",
"demo.launch()\n"
]
},
{
"cell_type": "code",
"execution_count": 70,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Closing server running on port: 7868\n"
]
}
],
"source": [
"demo.close()\n"
]
},
{
"cell_type": "code",
"execution_count": 71,
"metadata": {},
"outputs": [],
"source": [
"from nbdev.export import nb_export\n",
"\n",
"nb_export(\"app.ipynb\", lib_path=\".\", name=\"app\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "hf",
"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": "3.8.13"
},
"orig_nbformat": 4,
"vscode": {
"interpreter": {
"hash": "66e5af1d4a3a75efffc7cd5a7f382675fc3ac06b0697676e06fa85c907378a99"
}
}
},
"nbformat": 4,
"nbformat_minor": 2
}