Llama: differenze tra le versioni
(Creata pagina con "Modello Open Source rilasciato da Meta nel Febbraio 2023. La prima versione contava 7, 13, 33 e 65 B di parametri, === Links === * [https://arxiv.org/abs/2302.13971 LLama 1 (27 Feb 2023)] * [https://arxiv.org/abs/2302.13971 LLama 2]") |
Nessun oggetto della modifica |
||
| Riga 1: | Riga 1: | ||
Modello Open Source rilasciato da [[Meta]] nel Febbraio 2023. | Modello Open Source rilasciato da [[Meta]] nel Febbraio 2023. | ||
La prima versione contava 7, 13, 33 e 65 B di parametri, | La prima versione contava 7, 13, 33 e 65 B di parametri. | ||
"The capabilities of LLMs are remarkable considering the seemingly straightforward nature of the training methodology. Auto-regressive transformers are pretrained on an extensive corpus of self-supervised data, followed by alignment with human preferences via techniques such as Reinforcement Learning with Human Feedback (RLHF). Although the training methodology is simple, high computational requirements have limited the development of LLMs to a few players. There have been public releases of pretrained LLMs (such as BLOOM (Scao et al., 2022), LLaMa-1 (Touvron et al., 2023), and Falcon (Penedo et al., 2023)) that match the performance of closed pretrained competitors like GPT-3 (Brown et al., 2020) and Chinchilla (Hoffmann et al., 2022), but none of these models are suitable substitutes for closed “product” LLMs, such as ChatGPT, BARD, and Claude. These closed product LLMs are heavily fine-tuned to align with human preferences, which greatly enhances their usability and safety. This step can require significant costs in compute and human annotation, and is often not transparent or easily reproducible, limiting progress within the community to advance AI alignment research." da "Llama 2: Open Foundation and Fine-Tuned Chat Models" | |||
Versione delle 10:13, 3 mar 2024
Modello Open Source rilasciato da Meta nel Febbraio 2023. La prima versione contava 7, 13, 33 e 65 B di parametri.
"The capabilities of LLMs are remarkable considering the seemingly straightforward nature of the training methodology. Auto-regressive transformers are pretrained on an extensive corpus of self-supervised data, followed by alignment with human preferences via techniques such as Reinforcement Learning with Human Feedback (RLHF). Although the training methodology is simple, high computational requirements have limited the development of LLMs to a few players. There have been public releases of pretrained LLMs (such as BLOOM (Scao et al., 2022), LLaMa-1 (Touvron et al., 2023), and Falcon (Penedo et al., 2023)) that match the performance of closed pretrained competitors like GPT-3 (Brown et al., 2020) and Chinchilla (Hoffmann et al., 2022), but none of these models are suitable substitutes for closed “product” LLMs, such as ChatGPT, BARD, and Claude. These closed product LLMs are heavily fine-tuned to align with human preferences, which greatly enhances their usability and safety. This step can require significant costs in compute and human annotation, and is often not transparent or easily reproducible, limiting progress within the community to advance AI alignment research." da "Llama 2: Open Foundation and Fine-Tuned Chat Models"