user avatar
Andrew Drozdov
@mrdrozdov
Search and Agents @ Databricks
SF
mrdrozdov.github.io
Joined August 2010
1,993
Following
3,563
Followers
Posts
  • Pinned
    user avatar
    Andrew Drozdov
    @mrdrozdov
    Mar 5
    Today, we're sharing ๐ŸŒ Knowledge Agents from Reinforcement Learning (KARL) ๐ŸŒ We trained an agent that excels on challenging grounded reasoning tasks. KARL matches Sonnet 4.5 quality at a fraction of the cost, and with test-time scaling reaches Opus 4.6 levels. This was a fun
    user avatar
    Jonathan Frankle
    @jefrankle
    Mar 5
    Meet KARL, an RL'd model for document-centric tasks at frontier quality and open source cost/speed. Great for @databricks customers and scientists (77-page tech report!) As usual, this isn't just one model - it's an RL assembly line to churn out models for us and our customers ๐Ÿงต
    14K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Feb 21, 2024
    ๐ŸŒŸ PhD Thesis Defended ๐ŸŒŸ 1๏ธโƒฃ Title: Unlocking Natural Language Generalization through Adaptive Retrieval-based Methods 2๏ธโƒฃ Joining Databricks as a Research Scientist w. focus on generative retrieval / RAG 3๏ธโƒฃ New Blog Post: Advice for PhD Students mrdrozdov.github.io/blog/2024/adviโ€ฆ
    65K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Sep 30, 2022
    ๐Ÿšจ New preprint! ๐Ÿšจ We refine least-to-most prompting and achieve sota on CFQ (95% accuracy), outperforming previous fully supervised methods. Joint first author work with the formidable Nathanael Schรคrli.
    user avatar
    AK
    @_akhaliq
    Sep 30, 2022
    Compositional Semantic Parsing with Large Language Models abs: arxiv.org/abs/2209.15003
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Feb 28, 2024
    Replying to @jxmnop
    Fun fact. DPO author is also bulgarian. :)
    31K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Dec 2, 2022
    If you're applying for graduate school in CS / NLP, then definitely look at UMass! There's a vibrant NLP community with multiple incredible labs across many departments (ML, NLP, IR, RL, and more). I would strongly recommend UMass for MS or PhD. Happy to chat if interested!
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Oct 6, 2022
    โœจ Accepted at Findings of EMNLP 2022: You can't pick your neighbors, or can you? When and how to rely on retrieval in the kNN-LM โœจ We improve kNN-LM by incorporating retrieval quality. Joint work with @shufan_wang_, @Negin_Rahimi, @andrewmccallum, @HamedZamani, @MohitIyyer
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Aug 2, 2024
    Want to train and deploy large neural nets? Make them fast and robust? Mosaic x @databricks is hiring. We're especially looking for research engineers (at all levels). Send me a DM or email if you're interested. Happy to chat more about what this job is like.
    19K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Oct 24, 2023
    โœจ New Paper โœจ Deep dive on demonstrations to enhance LLM-based passage ranking ๐Ÿš€ insights for pointwise ranking using query likelihood ๐Ÿš€
    Paper page - PaRaDe: Passage Ranking using Demonstrations with Large Language Models
    From huggingface.co
    24K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Aug 28, 2024
    user avatar
    Tanishq Mathew Abraham, Ph.D.
    @iScienceLuvr
    Aug 28, 2024
    The Mamba in the Llama: Distilling and Accelerating Hybrid Models abs: arxiv.org/abs/2408.15237 code: github.com/jxiw/MambaInLlโ€ฆ "We demonstrate that it is feasible to distill large Transformers into linear RNNs by reusing the linear projection weights from attention layers with
    14K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Jul 18, 2018
    Starting a PhD in Computer Science at UMass-Amherst this fall. Focus will be on natural language processing and deep learning. Looking forward to reading even more papers than I do now, maybe even write a few. ๐Ÿ“š๐Ÿ“–โœ๏ธ๐Ÿง
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Apr 4, 2019
    Now with paper link: arxiv.org/abs/1904.02142 And code: github.com/iesl/diora New results on unsupervised parsing: +6.5 F1 compared to ON-LSTM (2019), +6 F1 compared to PRLG (2011).
    user avatar
    Andrew Drozdov
    @mrdrozdov
    Feb 22, 2019
    The Deep Inside-Outside Autoencoders have been accepted as a long paper at #NAACL2019 Unsupervised parsing and constituent representation with amazing co-authors @pat_verga Mohit Yadav @MohitIyyer @andrewmccallum
    arXiv logo
    arxiv.org
    Unsupervised Latent Tree Induction with Deep Inside-Outside...
    We introduce deep inside-outside recursive autoencoders (DIORA), a fully-unsupervised method for discovering syntax that simultaneously learns representations for constituents within the induced...
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Aug 23, 2023
    You can't win at #EMNLP2023. Paper 1: Reviewer complains we focus too much on a GPT-3 based model. How about performance on open source baselines? Paper 2: Reviewer complains we focus too much on open source baselines. Would this work for GPT-3?
    16K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Sep 25, 2024
    synthetic data creation has recently has a paradigm shift. it's no longer just about reducing your data annotation costs. the real benefit is creating data that simply would never be naturally occurring.
    user avatar
    Sasha Rush
    @srush_nlp
    Sep 25, 2024
    Long-context is central to models like OpenAI o1, but rare to see in natural data. Extension methods grow context by post-training open LLMs. A tutorial and controlled study of this area of long-context extension. arxiv.org/abs/2409.12181 youtu.be/dc4chADushM
    7.7K
  • user avatar
    Andrew Drozdov
    @mrdrozdov
    Aug 11, 2017
    Importance (and controversy) of deep learning in IR highlighted in a recent-ish slide from Chris Manning. nlp.stanford.edu/manning/talks/โ€ฆ