Arxiv Papers

The paper introduces LLaVA-Plus, a multimodal assistant trained using an end-to-end approach that expands the capabilities of large multimodal models. LLaVA-Plus outperforms previous models and enables new scenarios by actively engaging with users throughout the interaction. https://arxiv.org/abs//2311.05437 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers...

The paper introduces LLaVA-Plus, a multimodal assistant trained using an end-to-end approach that expands the capabilities of large multimodal models. LLaVA-Plus outperforms previous models and enables new scenarios by actively engaging with users throughout the interaction. https://arxiv.org/abs//2311.05437 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers...

The paper introduces a novel thought prompting approach called "Everything of Thoughts" (XOT) that enhances the capabilities of Large Language Models (LLMs) by incorporating external domain knowledge. XOT outperforms existing approaches in solving complex problems across different domains. https://arxiv.org/abs//2311.04254 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www...

The paper introduces a novel thought prompting approach called "Everything of Thoughts" (XOT) that enhances the capabilities of Large Language Models (LLMs) by incorporating external domain knowledge. XOT outperforms existing approaches in solving complex problems across different domains. https://arxiv.org/abs//2311.04254 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www...

The paper proposes a programmatic framework called RuLES for measuring the rule-following ability of Large Language Models (LLMs). It identifies attack strategies and evaluates the vulnerability of different LLMs to adversarial inputs. https://arxiv.org/abs//2311.04235 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple....

The paper proposes a programmatic framework called RuLES for measuring the rule-following ability of Large Language Models (LLMs). It identifies attack strategies and evaluates the vulnerability of different LLMs to adversarial inputs. https://arxiv.org/abs//2311.04235 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple....

The paper proposes a cascaded I2VGen-XL approach for video synthesis that improves semantic accuracy, clarity, and spatio-temporal continuity by decoupling factors and using static images as guidance. The approach enhances details and resolution and is effective on diverse data. https://arxiv.org/abs//2311.04145 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_pa...

The paper proposes a cascaded I2VGen-XL approach for video synthesis that improves semantic accuracy, clarity, and spatio-temporal continuity by decoupling factors and using static images as guidance. The approach enhances details and resolution and is effective on diverse data. https://arxiv.org/abs//2311.04145 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_pa...

S-LoRA is a system designed for scalable serving of many task-specific fine-tuned models. It improves throughput and increases the number of served adapters, enabling large-scale customized fine-tuning services. https://arxiv.org/abs//2311.03285 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-pap...

S-LoRA is a system designed for scalable serving of many task-specific fine-tuned models. It improves throughput and increases the number of served adapters, enabling large-scale customized fine-tuning services. https://arxiv.org/abs//2311.03285 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-pap...

This paper investigates how effectively transformer models can learn new tasks in-context, both within and outside their pretraining distribution. Results show that while transformers excel at learning tasks within their pretraining data, they struggle with out-of-domain tasks. https://arxiv.org/abs//2311.00871 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_pap...

This paper investigates how effectively transformer models can learn new tasks in-context, both within and outside their pretraining distribution. Results show that while transformers excel at learning tasks within their pretraining data, they struggle with out-of-domain tasks. https://arxiv.org/abs//2311.00871 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_pap...

The paper proposes an enhanced approach, called OCTHaGOn, for solving black-box global optimization problems by approximating nonlinear constraints using machine learning models and incorporating adaptive sampling and robust optimization techniques. The approach is tested on 81 instances and shows improvements in solution feasibility and optimality compared to existing methods. https://arxiv.org/a...

The paper proposes an enhanced approach, called OCTHaGOn, for solving black-box global optimization problems by approximating nonlinear constraints using machine learning models and incorporating adaptive sampling and robust optimization techniques. The approach is tested on 81 instances and shows improvements in solution feasibility and optimality compared to existing methods. https://arxiv.org/a...

The paper explores simplifying the standard transformer block by removing various components without sacrificing training speed. Experimental results show that the simplified transformers achieve comparable performance with faster training throughput and fewer parameters. https://arxiv.org/abs//2311.01906 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Ap...

The paper explores simplifying the standard transformer block by removing various components without sacrificing training speed. Experimental results show that the simplified transformers achieve comparable performance with faster training throughput and fewer parameters. https://arxiv.org/abs//2311.01906 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Ap...

https://arxiv.org/abs//2310.17688 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers

https://arxiv.org/abs//2310.17688 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers

FlashDecoding++ is a fast Large Language Model (LLM) inference engine that addresses challenges in LLM acceleration, such as synchronized softmax update, under-utilized computation of flat GEMM, and static dataflow. It achieves significant speedups compared to existing implementations. https://arxiv.org/abs//2311.01282 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@a...

The paper presents Distil-Whisper, a smaller variant of the Whisper speech recognition model, achieved through distillation and pseudo-labelling. Distil-Whisper is faster and has fewer parameters while maintaining performance and robustness. Code and models are publicly available. https://arxiv.org/abs//2311.00430 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_...

The paper presents Distil-Whisper, a smaller variant of the Whisper speech recognition model, achieved through distillation and pseudo-labelling. Distil-Whisper is faster and has fewer parameters while maintaining performance and robustness. Code and models are publicly available. https://arxiv.org/abs//2311.00430 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_...

This paper explores different approaches to rendering text in pixel-based language models. The authors find that using character bigram rendering improves performance on sentence-level tasks without compromising performance on token-level or multilingual tasks. This approach also allows for a more compact model with comparable performance to larger models. https://arxiv.org/abs//2311.00522 YouTube...

This paper explores different approaches to rendering text in pixel-based language models. The authors find that using character bigram rendering improves performance on sentence-level tasks without compromising performance on token-level or multilingual tasks. This approach also allows for a more compact model with comparable performance to larger models. https://arxiv.org/abs//2311.00522 YouTube...

Generative AI models can produce outputs that challenge or exceed human capabilities, but they still make basic errors in understanding. This paradox is due to a divergence in the configuration of intelligence between models and humans, where models can generate expert-level outputs without fully understanding them. Experimental results show that models outperform humans in generation but fall sho...

Generative AI models can produce outputs that challenge or exceed human capabilities, but they still make basic errors in understanding. This paradox is due to a divergence in the configuration of intelligence between models and humans, where models can generate expert-level outputs without fully understanding them. Experimental results show that models outperform humans in generation but fall sho...