The map of machine-learning research.

# your coding agent, with the nisa-cli skill installed
$ nisa search --query "speculative decoding" \
    --query "draft-then-verify decoding" --date-from 2024 --max-results 5 \
    | jq -r '.papers[] | "\(.arxiv_id)  \(.year)  \(.citation_count)  \(.title)"'
2401.07851  2024   76  Unlocking Efficiency in Large Language Model Inference: A Comprehensive…
2404.16710  2024   69  LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding
2401.10774  2024  201  Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads
2401.15077  2024   89  EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty
2402.12374  2024   40  Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding
$ nisa related 2401.10774 --n 3 \
    | jq -r '.papers[] | "\(.arxiv_id)  \(.year)  \(.citation_count)  \(.title)"'
2302.01318  2023   337  Accelerating Large Language Model Decoding with Speculative Sampling
2401.15077  2024    89  EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty
2303.08774  2023  14961  GPT-4 Technical Report
$ nisa excerpts 2401.10774 --query "acceptance" --max 1
{
  "arxiv_id": "2401.10774", "matched_terms": ["acceptance"], "full_text_indexed": true,
  "excerpts": [{ "index": 0, "text": "…we propose several extensions that improve or expand
    the utility of MEDUSA, including a self-distillation … and a typical acceptance scheme
    to boost the acceptance…" }]
}