Meta's LLaMA: Open and Efficient Foundation Language Models скачать в хорошем качестве

Meta's LLaMA: Open and Efficient Foundation Language Models

LLaMA is a collection of foundation language models from 7B to 65B parameters trained by Meta. They have been trained on 1.4T tokens from publicly available datasets exclusively. LLaMA-13B is better than GPT-3 while being more than 10 times smaller, and LLaMA-65B performs comparable to Chinchilla-70B and PaLM-540B. Here is the agenda for this video: 00:00:00 What is LLaMA? 00:01:40 How is LLaMA pretrained? 00:03:24 How does the architecture of LLaMA models look like? 00:04:20 How does LLaMA perform on Common Sense Reasoning, Closed-book QA, Reading Comprehension, Mathematical reasoning, Code generation and Massive Multitask Language Understanding (MMLU)? 00:12:25 Bias, Toxicity and Misinformation analysis of LLaMA 00:17:52 Examples of generations from LLaMA-65B For more details, please look at https://arxiv.org/pdf/2302.13971.pdf https://github.com/facebookresearch/l... Touvron, Hugo, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière et al. "Llama: Open and efficient foundation language models." arXiv preprint arXiv:2302.13971 (2023).