Skip to content

kei-saito-research/nrr-ime

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

49 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NRR-IME: Interface Decomposition for Stable State Transitions on Stateless LLM APIs

NRR-IME provides an interface-structure package for ambiguity-preserving inference under stateless LLM API constraints. The engineering objective is to reduce premature commitment in LLM decoding and limit downstream rework caused by semantic collapse. The control policy is defer vs commit under explicit conditions: maintain compatible alternatives while evidence is weak, then commit when action boundaries are clear. This repository includes the current manuscript snapshot and reproducibility assets for benchmarking which interface decomposition yields stable state transitions, with explicit protocol constraints and condition-bounded claims.

Quick links

EN/JA query terms

  • early commitment = 早期確定
  • ambiguity-preserving inference = 曖昧性保持推論

Part of the Non-Resolution Reasoning (NRR) research program.

NRR Series Hub (Start here)

For the cross-paper map and current series links, start here:

NRR is not an anti-LLM framework. NRR does not replace standard LLM use. NRR optimizes when to commit and when to defer, under explicit conditions. Series numbering policy: paper3 is permanently skipped and never reused.

DOI

DOI

Manuscript Artifacts

Current manuscript snapshot:

  • manuscript/current/paper4-nrr-ime-v67.tex
  • manuscript/current/paper4-nrr-ime-v67.pdf
  • manuscript/current/fig1_design_space.png
  • manuscript/current/fig2_bangbang.png
  • manuscript/current/fig3_comparative.png
  • manuscript/current/fig4_stability.png
  • manuscript/current/checksums_sha256.txt

Publication posting timeline may differ by platform availability.

Repository Structure

nrr-ime/
|-- README.md
|-- LICENSE
|-- requirements.txt
|-- reproducibility.md
|-- manuscript/
|   `-- current/
|       |-- paper4-nrr-ime-v67.tex
|       |-- paper4-nrr-ime-v67.pdf
|       |-- fig1_design_space.png
|       |-- fig2_bangbang.png
|       |-- fig3_comparative.png
|       |-- fig4_stability.png
|       `-- checksums_sha256.txt
|-- experiments/
|   |-- experimental_data.json
|   |-- generate_figures.py
|   |-- paper4_crossmodel_v5.ipynb
|   |-- phase_comparison.py
|   `-- scaling_validation.py
`-- .gitignore

Quick Start

pip install -r requirements.txt
bash scripts/generate_manuscript_figures.sh
python3 experiments/phase_comparison.py
python3 experiments/scaling_validation.py

The bundled experiments/experimental_data.json is the merged Paper 4 experiment dataset used for the current manuscript line. It includes the 135 run-level records used by the repository comparison scripts. Full provider-side infrastructure logs are not bundled in this snapshot.

experiments/generate_figures.py regenerates the four manuscript figure PNGs from the bundled merged dataset. The stable wrapper writes to a temp output directory by default so the tracked review package remains unchanged.

The main experiment implementation used for this dataset is bundled as experiments/paper4_crossmodel_v5.ipynb.

Reproducibility

See reproducibility.md for fixed settings and artifact mapping.

Stable review-package entrypoints:

  • bash scripts/generate_manuscript_figures.sh
  • bash scripts/build_current_manuscript.sh
  • bash scripts/verify_current_package.sh

Related Repositories

Collaboration Style

I support written technical Q&A, concept clarification, and small evaluation design.

Typical flow:

  1. you send questions and context,
  2. I return a structured technical response,
  3. if needed, I provide an English-ready version for external sharing.

Scope: research interpretation and evaluation planning.
Out of scope: production integration, implementation outsourcing, ongoing operations, and SLA/deadline commitments.
Contact: kei.saito.research@gmail.com

License

CC BY 4.0. See LICENSE.