Headless Language Models: Learning without Predicting with Contrastive Weight Tying https://
arxiv.org/abs/2309.08351 @nthngdy Éric de la Clergerie @bensagot
RESEARCH
-
Headless Language Models: Learning Without Predicting
By
–
-
Draft & Verify: Lossless LLM Acceleration via Self-Speculative Decoding
By
–
Draft & Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding https://
arxiv.org/abs/2309.08168
Jun Zhang
Jue Wang
Huan Li
Lidan Shou
Ke Chen
Gang Chen
Sharad Mehrotra -
Data Pruning Strategies for Large Language Model Pretraining
By
–
When Less is More: Investigating Data Pruning for Pretraining LLMs at Scale https://
cohere.com/research/paper
s/when-less-is-more-investigating-data-pruning-for-pretraining-llms-at-scale-2023-09-08
… @maxdoesresearch @ahmetustun89 @luizapzbn @W4ngatang @mziizm @sarahookr -
Mixture of Experts Parameter Efficiency for Instruction Tuning
By
–
Pushing Mixture of Experts to the Limit: Extremely Parameter Efficient MoE for Instruction Tuning https://
cohere.com/research/paper
s/pushing-mixture-of-experts-to-the-limit-extremely-parameter-efficient-moe-for-instruction-tuning-2023-09-11
… @tedzadouri Ahmet Üstün @aahmadian_ @beyzaermis @acyr_l @sarahookr -
Software Portability Myth and Machine Learning Progress Implications
By
–
The Grand Illusion: The Myth of Software Portability and Implications for ML Progress https://
cohere.com/research/paper
s/the-grand-illusion-the-myth-of-software-portability-and-implications-for-ml-progress-2023-09-12
… @fraser_mince @dzungdinhh @jonas_kg @ProfNeilT @sarahookr -
Top NLP Research Papers September 2023 Curated by Cohere
By
–
Stay up-to-date with the latest research in NLP! Here are some of the top recent papers curated by the Cohere for AI community! https://
txt.cohere.com/top-nlp-papers
-september-2023/
…
Featuring the following papers: (thread) -
RAG Reranking: Enhancing Retrieval with Pinecone Cohere
By
–
RAG with reranking – leveraging @pinecone and @cohere Reranking is a retrieval technique that performs an additional step on top of retrieved results This step uses a separate model to rerank the results, making sure the most relevant ones surface to the top @jamescalam
-
Google Research Event Showcases AI Climate Education Healthcare Innovations
By
–
Research@ NYC has come to a close! The event highlighted how researchers at Google are exploring innovations in climate, education, generative AI, health, & more — important work they're doing at Google to help define the technology of today & tomorrow.
-
Advanced Math Solves Language Distance and Name Schema Problems
By
–
Extraordinary that sufficiently advanced math manages to cache obscure org charts while also solving distance-in-pronunciation for all words and languages simultaneously while also, no big deal, solving the problem of the human name schema that has bedeviled every program ever.
-
Exciting New AI Paradigm Developments and Opportunities
By
–
It is! Very exciting and fun to try though, especially with the new AI paradigm.