├── README.md
└── summarize_videos_clean.ipynb


/README.md:
--------------------------------------------------------------------------------
 1 | # GPT Summarizer
 2 | 
 3 | This notebook can be used to download videos from YouTube, transcribe locally with a distilled version of Whisper and then summarized by calling gpt-4-turbo
 4 | 
 5 | Enjoy!
 6 | 
 7 | PS: if you find it useful, please 🌟 the repo, thanks!
 8 | 
 9 | Obligatory blog self-promotion:
10 | 
11 | [👉 Cool machine learning tutorials](https://forecastegy.com/tags/machine-learning/)


--------------------------------------------------------------------------------
/summarize_videos_clean.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "code",
  5 |    "execution_count": null,
  6 |    "metadata": {},
  7 |    "outputs": [],
  8 |    "source": [
  9 |     "import json\n",
 10 |     "import re\n",
 11 |     "import pathlib\n",
 12 |     "import openai\n",
 13 |     "import os\n",
 14 |     "path = pathlib.Path(\"path_to_good_stuff\")"
 15 |    ]
 16 |   },
 17 |   {
 18 |    "cell_type": "code",
 19 |    "execution_count": null,
 20 |    "metadata": {
 21 |     "vscode": {
 22 |      "languageId": "plaintext"
 23 |     }
 24 |    },
 25 |    "outputs": [],
 26 |    "source": [
 27 |     "from yt_dlp import YoutubeDL\n",
 28 |     "\n",
 29 |     "URLS = ['channel, video, or playlist urls']\n",
 30 |     "with YoutubeDL(params={'format': '140', \"paths\": {\"home\": path.as_posix()}}\n",
 31 |     "               ) as ydl:\n",
 32 |     "    ydl.download(URLS)"
 33 |    ]
 34 |   },
 35 |   {
 36 |    "cell_type": "code",
 37 |    "execution_count": null,
 38 |    "metadata": {
 39 |     "vscode": {
 40 |      "languageId": "plaintext"
 41 |     }
 42 |    },
 43 |    "outputs": [],
 44 |    "source": [
 45 |     "import torch\n",
 46 |     "from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline\n",
 47 |     "\n",
 48 |     "model_id = \"distil-whisper/distil-large-v2\"\n",
 49 |     "torch_dtype = torch.float16\n",
 50 |     "device = \"cuda:0\"\n",
 51 |     "\n",
 52 |     "model = AutoModelForSpeechSeq2Seq.from_pretrained(\n",
 53 |     "    model_id, torch_dtype=torch_dtype, \n",
 54 |     "    low_cpu_mem_usage=True, use_safetensors=True, \n",
 55 |     "    use_flash_attention_2=False\n",
 56 |     "    ).to_bettertransformer()\n",
 57 |     "model.to(device)\n",
 58 |     "\n",
 59 |     "processor = AutoProcessor.from_pretrained(model_id)\n",
 60 |     "\n",
 61 |     "pipe = pipeline(\"automatic-speech-recognition\",\n",
 62 |     "                model=model,\n",
 63 |     "                tokenizer=processor.tokenizer,\n",
 64 |     "                feature_extractor=processor.feature_extractor,\n",
 65 |     "                torch_dtype=torch_dtype,\n",
 66 |     "                device=device)\n",
 67 |     "\n",
 68 |     "for fname in path.glob(\"*.m4a\"):\n",
 69 |     "    if fname.with_suffix(\".txt\").exists():\n",
 70 |     "        continue\n",
 71 |     "    \n",
 72 |     "    print(fname)\n",
 73 |     "    outputs = pipe(fname.as_posix(),\n",
 74 |     "                chunk_length_s=15,\n",
 75 |     "                batch_size=32,\n",
 76 |     "                return_timestamps=True)\n",
 77 |     "\n",
 78 |     "    text = outputs[\"text\"]\n",
 79 |     "    #chunks = [text[i:i+6000] for i in range(0, len(text), 6000)]\n",
 80 |     "\n",
 81 |     "    with open(fname.with_suffix(\".txt\"), \"w\") as f:\n",
 82 |     "        #for c in chunks:\n",
 83 |     "        #    f.write(c)\n",
 84 |     "        #    f.write(\"\\n\\n\\n\")\n",
 85 |     "        f.write(text)\n",
 86 |     "    "
 87 |    ]
 88 |   },
 89 |   {
 90 |    "cell_type": "code",
 91 |    "execution_count": null,
 92 |    "metadata": {},
 93 |    "outputs": [],
 94 |    "source": [
 95 |     "openai.api_key = os.getenv(\"OPENAI_API_KEY\")\n",
 96 |     "\n",
 97 |     "i = 0\n",
 98 |     "for fname in path.glob(\"*.txt\"):\n",
 99 |     "    if (path / \"summaries\" / fname.name).exists():\n",
100 |     "        continue\n",
101 |     "    if i > 4:\n",
102 |     "        break\n",
103 |     "    print(fname)\n",
104 |     "    transcript = fname.read_text().strip()\n",
105 |     "    title = re.sub(r'\\[.*?\\]', '', fname.stem).strip()\n",
106 |     "    \n",
107 |     "    \n",
108 |     "    system_msg = f\"\"\"Give me a detailed summary of the following transcript. The title is: {title}\"\"\"\n",
109 |     "    #print(system_msg)\n",
110 |     "    prompt = f\"\"\"TRANSCRIPT:\\n\\n{transcript}\"\"\"\n",
111 |     "\n",
112 |     "    completion = openai.ChatCompletion.create(\n",
113 |     "        model=\"gpt-4-1106-preview\",\n",
114 |     "        temperature=0.0,\n",
115 |     "        max_tokens=1000,\n",
116 |     "        messages=[\n",
117 |     "            {\"role\": \"system\", \"content\": system_msg},\n",
118 |     "            {\"role\": \"user\", \"content\": prompt},\n",
119 |     "        ],\n",
120 |     "    )\n",
121 |     "\n",
122 |     "    summary = completion['choices'][0].message['content']\n",
123 |     "\n",
124 |     "\n",
125 |     "    with open(path / \"summaries\" / fname.name, \"w\") as f:\n",
126 |     "        f.write(summary)\n",
127 |     "    i += 1    "
128 |    ]
129 |   }
130 |  ],
131 |  "metadata": {
132 |   "kernelspec": {
133 |    "display_name": "seo",
134 |    "language": "python",
135 |    "name": "python3"
136 |   },
137 |   "language_info": {
138 |    "codemirror_mode": {
139 |     "name": "ipython",
140 |     "version": 3
141 |    },
142 |    "file_extension": ".py",
143 |    "mimetype": "text/x-python",
144 |    "name": "python",
145 |    "nbconvert_exporter": "python",
146 |    "pygments_lexer": "ipython3",
147 |    "version": "3.9.16"
148 |   }
149 |  },
150 |  "nbformat": 4,
151 |  "nbformat_minor": 2
152 | }
153 | 


--------------------------------------------------------------------------------