Research map / Learning theory and algorithms / Combinatorial and search optimization
Intermediate N-Gramming: Deterministic and Fast N-Grams for Large N and Large Datasets
Ryan R. Curtin, Fred Lu, Edward Raff, Priyanka Ranade
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace hash-gramming (hg-fast) with Intergrams z=1 | top-k n-gram mining · C4 (k=10k) | runtime (s): 30811.2s -> 1183.9s | Improved |
| replace hg-fast with Intergrams z=1 | top-k n-gram mining · EMBER (k=100k) | runtime (s): 7162.6s -> 1413.7s | Improved |
| replace hg-fast with Intergrams z=1 | top-k n-gram mining · 1000gp (k=10k) | runtime (s): 7061.5s -> 1215.7s | Improved |
| increase oversampling z from 1 to 1.5 (ablation) | Intergrams · EMBER (k=100k) | Jaccard with true top-k: 0.71 -> 0.91 | Improved |
| increase oversampling z from 1.5 to 2 (ablation) | Intergrams · EMBER (k=100k) | Jaccard with true top-k: 0.91 -> 0.97 | Improved |
| increase oversampling z from 1 to 2 (ablation) | Intergrams · EMBER (k=100k) | speedup: 6.42x -> 5.13x | Hurt |
| increase oversampling z from 1 to 1.5 (ablation) | Intergrams · C4 (k=10k) | Jaccard with true top-k: 0.91 -> 1.0 | Improved |
| increase oversampling z from 1 to 2 (ablation) | Intergrams · C4 (k=10k) | speedup: 33.6x -> 26.5x | Hurt |
| increase oversampling z from 1 to 2 (ablation) | Intergrams · 1000gp (k=10k) | Jaccard with true top-k: 1.0 | No clear effect |
| add 2 MB large pages to hash-gramming (ablation) | hg-vanilla · EMBER (k=100k) | speedup: 1.01x -> 1.08x | Improved |
| add cuckoo hashing and other cumulative tricks to hash-gramming (ablation) | hg-vanilla · C4 (k=10k) | speedup: 1.01x -> 1.06x (hg-cuckoo), 1.29x (hg-fast) | Improved |
| use trie for exact-count pass in hash-gramming (ablation) | hash-gramming · 1000gp (k=10k) | speedup: 1.16x -> 1.24x | Improved |
Closest papers by meaning
- Learning In-context $n$-grams with Transformers: Sub-$n$-grams Are Near-Stationary Points (ICML 2025 · similarity 0.84)
- DNA-GPT: Divergent N-Gram Analysis for Training-Free Detection of GPT-Generated Text (ICLR 2024 · similarity 0.82)
- Weak-to-Strong Generalization Through the Data-Centric Lens (ICLR 2025 · similarity 0.81)
- Regress, Don't Guess: A Regression-like Loss on Number Tokens for Language Models (ICML 2025 · similarity 0.81)
- Cut Your Losses in Large-Vocabulary Language Models (ICLR 2025 · similarity 0.81)
- PEAKS: Selecting Key Training Examples Incrementally via Prediction Error Anchored by Kernel Similarity (ICML 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
