This is a fine-tuned LLM based on Llama 3.1 8B, trained off of the titles of top performing videos as of June 2025. Input a 2 sentence summary of what your video is about, and receive a well-worded title that should perform well against the YouTube algorithm. Huggingface.

• Ollama: https://github.com/ollama/ollama
• Unsloth: https://github.com/unslothai/unsloth
• Python: https://www.python.org/
• Youtube Transcript API: https://pypi.org/project/youtube-transcript-api/

• CPU: Intel Core i9-13900K
• RAM: 64 GB of DDR5
• GPU: NVIDIA RTX 4090 with 24 GB of VRAM
• Operating System: Ubuntu 22.04 LTS
• NVIDIA Driver Version: 550.67
• CUDA Version: 11.8

1. Create a python script to create a JSON formatted dataset of the top 1000 best-performing videos.
   1. Find dataset of top 1000 best performing videos (hopefully with video ID)
   2. Scrape (for educational purposes) the English transcript and the video title.
   3. Add to JSON object for singular video example

Input: Transcript (with no timestamps) of top-performing input video.
- "Hello guys! In this video, we will be ____"

Output: YouTube video title of same top-performing video.
- "INSANE 5 MINUTE HACK TO __"

2. Fine-tune using Unsloth and the dataset
3. Make highly optimized YouTube video titles based on any video transcript (w/no time stamps). Export model and publish to Huggingface.