1. X
  2. Sumit
Log inSign up
Sumit
10.6K posts
Sumit profile banner
@_reachsumit

Sumit

@_reachsumit
Senior ML Engineer @Meta | prev: @TikTok_us, @Amazon, @Samsung | UChicago Alum blog.reachsumit.com ๐Ÿ‡ฎ๐Ÿ‡ณโ†’๐Ÿ‡ฐ๐Ÿ‡ทโ†’๐Ÿ‡ฆ๐Ÿ‡บโ†’๐Ÿ‡จ๐Ÿ‡ฆโ†’๐Ÿ‡บ๐Ÿ‡ฒ
Seattle, WA
recsys.substack.com
Joined April 2010
511
Following
4,274
Followers
RepliesRepliesRepostsRepostsMediaMedia

Log in or sign up for X

See whatโ€™s happening and join the conversation

Continue with phone
or
Log in with username or email
TermsยทPrivacyยทCookiesยทAccessibilityยทAds Infoยทยฉ 2026 X Corp.
  • Pinned
    @_reachsumit
    Sumit
    @_reachsumit
    Oct 8, 2025
    In the final post of the Adaptive RAG series, we explore how to treat selective retrieval as a core, learned skill, moving from passive observation to active, intelligent decision-making.
    Image
    Teaching Models to Decide When to Retrieve: Adaptive RAG, Part 4
    From blog.reachsumit.com
  • @_reachsumit
    Sumit
    @_reachsumit
    11h
    I published Vol. 171 of "Top Information Retrieval Papers of the Week" on Substack. ๐Ÿ”— recsys.substack.com/p/eliminating-โ€ฆ
    Image
  • @_reachsumit
    Sumit
    @_reachsumit
    Aug 28
    PuRo-2B: Poor Lab's Qwen2-1.5B Trained on RTX 5090 within $5090 @openhonor et al. present an open, low-cost recipe for pretraining a 2B LLM on consumer RTX 5090 GPUs, matching Qwen2-1.5B at ~$6.9K. ๐Ÿ“ arxiv.org/abs/2608.27370 ๐Ÿ‘จ๐Ÿฝโ€๐Ÿ’ป github.com/thu-pacman/Purโ€ฆ ๐Ÿค— huggingface.co/collections/thโ€ฆ
    arXiv logo
    arxiv.org
    Puro-2B: Poor Lab's Qwen2-1.5B Trained on RTX 5090 within $5090
    Language model pretraining has become almost synonymous with prohibitive cost, placing it out of reach for much of the academic and open-source communities. Although strong open-source efforts...
  • @_reachsumit
    Sumit
    @_reachsumit
    Aug 28
    CorporateBench: Large-Scale Q&A Benchmarking with Temporal Knowledge Bases @srhm_ca et al. present an enterprise Q&A benchmark built from synthetic companies with consistent knowledge bases. ๐Ÿ“ arxiv.org/abs/2608.27391 ๐Ÿค— huggingface.co/datasets/epiq-โ€ฆ
    arXiv logo
    arxiv.org
    CorporateBench: Large-Scale Q&A Benchmarking with Temporal...
    LLMs are increasingly able to answer complex questions about enterprise-scale document collections. But evaluation is hard: companies don't want to share internal communications, and synthetic...
  • @_reachsumit
    Sumit
    @_reachsumit
    Aug 27
    Unifying Search and Recommendation with Dual-View Representation Learning in a Generative Paradigm @JZhao0109 et a. unifiy search and recommendation in a generative model via task prompts and dual CF/semantic views. ๐Ÿ“ doi.org/10.1145/3841470 ๐Ÿ‘จ๐Ÿฝโ€๐Ÿ’ป github.com/Polaris-JZ/Genโ€ฆ
    dl.acm.org
    Unifying Search and Recommendation with Dual-View Representation Learning in a Generative Paradigm...
    Recommender systems and personalized product search are critical components of modern online platforms, where recommender systems proactively surface relevant items while product search allows users...
Advertisement
Advertisement