Igor Melnyk
Arxiv Papers
Running out of time to catch up with new arXiv papers? We take the most impactful papers and present them as convenient podcasts. If you're a visual learner, we offer these papers in an engaging video format. Our service fills the gap between overly brief paper summaries and time-consuming full paper reads. You gain academic insights in a time-efficient, digestible format. Code behind this work: https://github.com/imelnyk/ArxivPapers
Where to listen?
Podcasts in the app Replaio Radio Coming soonPodcasts are coming to the app soon. Install now and be the first to see a whole new take on podcasts
Install for free
Android 5M+ downloads · 4.8 rating
iOS soon
Episodes
[short] SEQUOIA: Scalable, Robust, and Hardware-aware Speculative Decoding 20.02.2024 2:38
SEQUOIA is a scalable, robust, and hardware-aware algorithm for speculative decoding, improving inference speed for large language models on various hardware platforms. https://arxiv.org/abs//2402.12374 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://podcaster...
SEQUOIA: Scalable, Robust, and Hardware-aware Speculative Decoding 20.02.2024 29:44
SEQUOIA is a scalable, robust, and hardware-aware algorithm for speculative decoding, improving inference speed for large language models on various hardware platforms. https://arxiv.org/abs//2402.12374 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://podcaster...
[short] Reformatted Alignment 20.02.2024 2:53
REALIGN improves large language models' alignment with human values by reformatting instruction data responses, enhancing math reasoning, factuality, and readability without additional data or training techniques. https://arxiv.org/abs//2402.12219 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arx...
Reformatted Alignment 20.02.2024 21:16
REALIGN improves large language models' alignment with human values by reformatting instruction data responses, enhancing math reasoning, factuality, and readability without additional data or training techniques. https://arxiv.org/abs//2402.12219 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arx...
[short] PaLM2-VAdapter: Progressively Aligned Language Model Makes a Strong Vision-language Adapter 19.02.2024 2:43
The paper introduces PaLM2-VAdapter, a progressively aligned language model as a vision-language adapter, showing faster convergence, higher performance, and stronger scalability in multi-modal tasks with fewer parameters. https://arxiv.org/abs//2402.10896 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcas...
PaLM2-VAdapter: Progressively Aligned Language Model Makes a Strong Vision-language Adapter 19.02.2024 17:54
The paper introduces PaLM2-VAdapter, a progressively aligned language model as a vision-language adapter, showing faster convergence, higher performance, and stronger scalability in multi-modal tasks with fewer parameters. https://arxiv.org/abs//2402.10896 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcas...
[short] In Search of Needles in a 10M Haystack: Recurrent Memory Finds What LLMs Miss 19.02.2024 2:40
The paper introduces BABILong benchmark to evaluate transformer models on processing long documents. Fine-tuning GPT-2 with memory augmentations enables handling tasks with up to elements, a significant advancement. https://arxiv.org/abs//2402.10790 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv...
In Search of Needles in a 10M Haystack: Recurrent Memory Finds What LLMs Miss 19.02.2024 13:01
The paper introduces BABILong benchmark to evaluate transformer models on processing long documents. Fine-tuning GPT-2 with memory augmentations enables handling tasks with up to elements, a significant advancement. https://arxiv.org/abs//2402.10790 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv...
[short] DoRA: Weight-Decomposed Low-Rank Adaptation 18.02.2024 2:42
DoRA enhances fine-tuning by decomposing weights into magnitude and direction, improving learning capacity and stability over LoRA without additional inference costs, outperforming on various downstream tasks. https://arxiv.org/abs//2402.09353 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-paper...
DoRA: Weight-Decomposed Low-Rank Adaptation 18.02.2024 24:39
DoRA enhances fine-tuning by decomposing weights into magnitude and direction, improving learning capacity and stability over LoRA without additional inference costs, outperforming on various downstream tasks. https://arxiv.org/abs//2402.09353 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-paper...
[short] A Human-Inspired Reading Agent with Gist Memory of Very Long Contexts 17.02.2024 2:34
https://arxiv.org/abs//2402.09727 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
A Human-Inspired Reading Agent with Gist Memory of Very Long Contexts 17.02.2024 20:47
https://arxiv.org/abs//2402.09727 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
[short] On Limitations of the Transformer Architecture 17.02.2024 2:15
The paper explores the root causes of hallucinations in large language models, demonstrating the Transformer layer's limitations in composing functions for tasks like genealogy identification. https://arxiv.org/abs//2402.08164 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id169247601...
On Limitations of the Transformer Architecture 17.02.2024 21:08
The paper explores the root causes of hallucinations in large language models, demonstrating the Transformer layer's limitations in composing functions for tasks like genealogy identification. https://arxiv.org/abs//2402.08164 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id169247601...
[short] Chain-of-Thought Reasoning Without Prompting 16.02.2024 2:36
https://arxiv.org/abs//2402.10200 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
Chain-of-Thought Reasoning Without Prompting 16.02.2024 21:06
https://arxiv.org/abs//2402.10200 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
[short] How to Train Data-Efficient LLMs 16.02.2024 2:32
Study explores data-efficient pre-training methods for large language models. ASK-LLM assesses data quality, DENSITY sampling selects diverse samples. Both outperform full-data training. https://arxiv.org/abs//2402.09668 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:...
How to Train Data-Efficient LLMs 16.02.2024 22:56
Study explores data-efficient pre-training methods for large language models. ASK-LLM assesses data quality, DENSITY sampling selects diverse samples. Both outperform full-data training. https://arxiv.org/abs//2402.09668 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:...
[short] Data Engineering for Scaling Language Models to 128K Context 16.02.2024 2:34
Study explores continual pretraining for scaling language models' context lengths to 128K, emphasizing data engineering's importance in achieving optimal performance and closing the gap to top models. https://arxiv.org/abs//2402.10171 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers...
Data Engineering for Scaling Language Models to 128K Context 16.02.2024 18:55
Study explores continual pretraining for scaling language models' context lengths to 128K, emphasizing data engineering's importance in achieving optimal performance and closing the gap to top models. https://arxiv.org/abs//2402.10171 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers...
[short] Premise Order Matters in Reasoning with Large Language Models 15.02.2024 2:33
https://arxiv.org/abs//2402.08939 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
Premise Order Matters in Reasoning with Large Language Models 15.02.2024 14:56
https://arxiv.org/abs//2402.08939 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
[short] HiRE: High Recall Approximate Top-k Estimation for Efficient LLM Inference 15.02.2024 2:21
HiRE introduces a method to reduce memory-bound autoregressive decoding with Large Language Models by efficiently predicting and computing top rows/columns, improving latency on accelerators. https://arxiv.org/abs//2402.09360 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spo...
HiRE: High Recall Approximate Top-k Estimation for Efficient LLM Inference 15.02.2024 27:16
HiRE introduces a method to reduce memory-bound autoregressive decoding with Large Language Models by efficiently predicting and computing top rows/columns, improving latency on accelerators. https://arxiv.org/abs//2402.09360 YouTube: https://www.youtube.com/@ArxivPapers TikTok: https://www.tiktok.com/@arxiv_papers Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016 Spo...
[short] Transformers Can Achieve Length Generalization But Not Robustly 15.02.2024 2:07
https://arxiv.org/abs//2402.09371 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
Similar podcasts
Listen to the Arxiv Papers podcast in Replaio
Radio and podcasts in one app - free, with no sign-up. Install today and do not miss the launch
Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.
