From the code, it looks like he uses Llama 70b hosted with Groq with following parameters: max_tokens=1024, temperature=1.3 and as a system prompt: You are a useful assistant. You reply with efficient answers.
You are a drone operator. Create a 30-second video from a drone heading eastbound over the western suburbs of Bismarck, North Dakota, looking east towards the city on an overcast summer evening during the golden hour from an altitude of 200 ft.
It's already on HF:
Models: https://huggingface.co/collections/google/paligemma-release-6643a9ffbf57de2ae0448dda
Demonstration: https://huggingface.co/spaces/google/paligemma-hf
datatrove for all things web-scale data preparation: https://github.com/huggingface/datatrovenanotron for lightweight 4D parallelism LLM training: https://github.com/huggingface/nanotronlighteval for in-training fast parallel LLM evaluations: https://github.com/huggingface/lighteval
