Best Large Language Models - Page 6

Chinchilla is a large language model. Chinchilla uses the same compute budget as Gopher but with 70B parameters and 4× more more data. Chinchilla uniformly and significantly outperforms Gopher (280B), GPT-3 (175B), Jurassic-1 (178B), and Megatron-Turing NLG (530B) on a large range of downstream evaluation tasks. This also means that Chinchilla uses substantially less compute for fine-tuning and inference, greatly facilitating downstream usage. As a highlight, Chinchilla reaches a state-of-the-art average accuracy of 67.5% on the MMLU benchmark, greater than a 7% improvement over Gopher.

Information overload is a major obstacle to scientific progress. The explosive growth in scientific literature and data has made it ever harder to discover useful insights in a large mass of information. Today scientific knowledge is accessed through search engines, but they are unable to organize scientific knowledge alone. Galactica is a large language model that can store, combine and reason about scientific knowledge. We train on a large scientific corpus of papers, reference material, knowledge bases and many other sources. We outperform existing models on a range of scientific tasks. On technical knowledge probes such as LaTeX equations, Galactica outperforms the latest GPT-3 by 68.2% versus 49.0%. Galactica also performs well on reasoning, outperforming Chinchilla on mathematical MMLU by 41.3% to 35.7%, and PaLM 540B on MATH with a score of 20.4% versus 8.8%.

Significant advancements in the field of natural language processing, understanding, and generation have been achieved through the expansion of large language models. This study introduces a system which utilizes Ascend 910 AI processors and the MindSpore framework to train a language model with over a trillion parameters, specifically 1.085T, named PanGu-{\Sigma}. This model, which builds upon the foundation laid by PanGu-{\alpha}, takes the traditionally dense Transformer model and transforms it into a sparse one using a concept known as Random Routed Experts (RRE). The model was efficiently trained on a dataset of 329 billion tokens using a technique called Expert Computation and Storage Separation (ECSS), leading to a 6.3-fold increase in training throughput via heterogeneous computing. Experimentation indicates that PanGu-{\Sigma} sets a new standard in zero-shot learning for various downstream Chinese NLP tasks.

OpenELM is an open-source language model family developed by Apple. It uses a layer-wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, leading to enhanced accuracy compared to existing open language models of similar size. OpenELM is trained on publicly available datasets and achieves state-of-the-art performance for its size.

LTM-2-mini is a 100M token context model: LTM-2-mini. 100M tokens equals ~10 million lines of code or ~750 novels. For each decoded token, LTM-2-mini’s sequence-dimension algorithm is roughly 1000x cheaper than the attention mechanism in Llama 3.1 405B1 for a 100M token context window. The contrast in memory requirements is even larger – running Llama 3.1 405B with a 100M token context requires 638 H100s per user just to store a single 100M token KV cache.2 In contrast, LTM requires a small fraction of a single H100’s HBM per user for the same context.