Reposition as NGramStore with compatible engine APIs and working quickstarts

.github/ISSUE_TEMPLATE/benchmark-result.yml

@@ -1,5 +1,5 @@
 name: Benchmark result
-description: Share a reproducible me_fasttext benchmark, replication, or negative result
+description: Share a reproducible NGramStore benchmark, replication, or negative result
 title: "[Benchmark] "
 labels:
   - benchmark
@@ -46,7 +46,7 @@ body:
     id: commit
     attributes:
       label: Git commit
-      description: The me_fasttext commit used for the run.
+      description: The NGramStore commit used for the run.
       placeholder: "abcdef1"
     validations:
       required: true
@@ -125,7 +125,7 @@ body:
         | Method | Memory | Load time | Storage | Query p50 | Query p95 | OOV coverage | Quality |
         | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
         | Original FastText | | | | | | | |
-        | me_fasttext | | | | | | | |
+        | NGramStore | | | | | | | |
     validations:
       required: true
   - type: textarea
@@ -163,15 +163,15 @@ body:
     id: citation_fit
     attributes:
       label: Citation or reuse fit
-      description: If this result supports a paper, report, deployment note, or comparison table, describe how me_fasttext is used.
+      description: If this result supports a paper, report, deployment note, or comparison table, describe how NGramStore is used.
       placeholder: |
         Example: used as a compact lexical embedding layer before dense reranking;
         compared against hash-bucketed FastText for memory and OOV recall.
   - type: textarea
     id: notes
     attributes:
       label: Notes
-      description: Anything surprising, missing, hard to reproduce, or unfavorable to me_fasttext.
+      description: Anything surprising, missing, hard to reproduce, or unfavorable to NGramStore.
   - type: checkboxes
     id: disclosure
     attributes:

.github/ISSUE_TEMPLATE/compression-comparison.yml

@@ -1,5 +1,5 @@
 name: Compression comparison
-description: Compare me_fasttext with sparse, quantized, or compact-serving baselines
+description: Compare NGramStore with sparse, quantized, or compact-serving baselines
 title: "[Compression] "
 labels:
   - benchmark
@@ -10,16 +10,17 @@ body:
         Use this for focused comparisons against model-compression,
         sparse-lookup, quantized embedding, or compact-serving baselines.
 
-        Please keep the claim narrow. me_fasttext is a compact lexical
-        embedding artifact, not a general LLM compression framework.
+        NGramStore is a learned n-gram vector storage/query backend.
+        Name the actual table and exact/approximate mode; separate storage,
+        query costs and downstream quality. FastText and Engram are applications.
   - type: dropdown
     id: comparison_type
     attributes:
       label: Comparison type
       options:
-        - Original FastText vs me_fasttext compact .z
-        - Quantized FastText vs me_fasttext compact .z
-        - Hash-free uncompressed vs compact export
+        - Dense or raw mmap vs native n-gram store
+        - Quantized table vs lossless native store
+        - DATrie/structural contribution ablation
         - Sparse or table-compression baseline
         - Lexical search or retrieval baseline
         - Dense embedding or reranker prefilter comparison
@@ -42,7 +43,7 @@ body:
     id: commit
     attributes:
       label: Git commit
-      description: The me_fasttext commit used for this run.
+      description: The NGramStore commit used for this run.
       placeholder: "abcdef1"
     validations:
       required: true
@@ -87,10 +88,10 @@ body:
       description: List all compared methods and explain omissions.
       placeholder: |
         compared:
-          - Original FastText:
+          - Dense resident / raw mmap:
           - Quantized FastText:
-          - Hash-free uncompressed:
-          - me_fasttext compact .z:
+          - Identity-layout native control:
+          - NGramStore native table/index/metadata:
         omitted:
     validations:
       required: true
@@ -126,7 +127,7 @@ body:
         | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
         | Original FastText | | | | | | | |
         | Quantized FastText | | | | | | | |
-        | me_fasttext .z | | | | | | | |
+        | NGramStore native | | | | | | | |
     validations:
       required: true
   - type: textarea

.github/ISSUE_TEMPLATE/config.yml

@@ -1,11 +1,11 @@
 blank_issues_enabled: true
 contact_links:
   - name: Evaluator quickstart
-    url: https://github.com/initial-d/me_fasttext/blob/main/docs/evaluator_quickstart.md
+    url: https://github.com/initial-d/ngram-store/blob/main/docs/evaluator_quickstart.md
     about: Start here if you want to run a first public-corpus benchmark.
   - name: Benchmark protocol
-    url: https://github.com/initial-d/me_fasttext/blob/main/docs/benchmark_protocol.md
+    url: https://github.com/initial-d/ngram-store/blob/main/docs/benchmark_protocol.md
     about: Use this for the fuller memory, latency, OOV, and quality report shape.
   - name: Outreach notes
-    url: https://github.com/initial-d/me_fasttext/blob/main/docs/outreach.md
+    url: https://github.com/initial-d/ngram-store/blob/main/docs/outreach.md
     about: Use this when adding the project to lists or sharing it with maintainers.

.github/ISSUE_TEMPLATE/evaluation-question.yml

@@ -1,5 +1,5 @@
 name: Evaluation question
-description: Ask about building, benchmarking, reproducing, or interpreting me_fasttext
+description: Ask about building, benchmarking, reproducing, or interpreting NGramStore
 title: "[Question] "
 labels:
   - question

.github/ISSUE_TEMPLATE/showcase.yml

@@ -7,7 +7,7 @@ body:
   - type: markdown
     attributes:
       value: |
-        Thanks for helping other readers understand where `me_fasttext` fits.
+        Thanks for helping other readers understand where `NGramStore` fits.
 
         Use this template for citations, public list entries, comparison notes,
         integration experiments, deployment notes, or concrete use cases. Short
@@ -35,7 +35,7 @@ body:
     id: summary
     attributes:
       label: Summary
-      description: Briefly describe how `me_fasttext` was used, cited, compared, or listed.
+      description: Briefly describe how `NGramStore` was used, cited, compared, or listed.
       placeholder: |
         Example: Used as a compact lexical embedding baseline before dense
         reranking on an entity-heavy product-title retrieval corpus.
@@ -82,7 +82,7 @@ body:
       label: Suggested public wording
       description: Optional wording maintainers can quote in README, docs, release notes, or distribution logs.
       placeholder: |
-        Example: me_fasttext was useful as a compact lexical embedding layer
+        Example: NGramStore was used as a frozen n-gram memory backend
         for OOV-heavy candidate generation before dense reranking.
   - type: checkboxes
     id: disclosure

.zenodo.json

@@ -1,14 +1,18 @@
 {
-  "title": "me_fasttext: Memory-Efficient FastText with Exact Trie N-Gram IDs and mmap Serving",
+  "title": "NGramStore: Compression, Storage and Exact Lookup for Learned N-Gram Vectors",
   "creators": [
     {
       "name": "Du, Yimin",
       "affiliation": "Independent Researcher"
     }
   ],
-  "description": "A FastText-derived C++ research prototype that replaces hash-bucketed character n-gram ids with exact trie-backed ids, then exports a compact mmap serving index through conservative n-gram row sharing and mark-compact layout.",
+  "description": "Formerly me_fasttext. A native compression, storage and exact-query engine for frozen learned n-gram vectors, with registered DATrie/mark-compact planning and FastText/Engram application integrations. Lossless and approximate modes are reported separately; the original FastText paper retains its title and attribution.",
   "license": "MIT",
   "keywords": [
+    "n-gram memory",
+    "vector compression",
+    "Engram",
+    "LLM memory",
     "fastText",
     "word embeddings",
     "subword embeddings",
@@ -28,7 +32,7 @@
       "resource_type": "publication-preprint"
     },
     {
-      "identifier": "https://github.com/initial-d/me_fasttext",
+      "identifier": "https://github.com/initial-d/ngram-store",
       "relation": "isIdenticalTo",
       "scheme": "url",
       "resource_type": "software"

CITATION.cff

@@ -1,13 +1,16 @@
 cff-version: 1.2.0
-message: "If you use me_fasttext or the accompanying method, please cite the arXiv paper."
-title: "Memory-Efficient FastText: A Comprehensive Approach Using Double-Array Trie Structures and Mark-Compact Memory Management"
+message: "NGramStore was formerly me_fasttext. Cite the original paper for its method and the exact software commit/report for later engine work."
+title: "NGramStore: Compression, Storage and Exact Lookup for Learned N-Gram Vectors"
 authors:
   - family-names: Du
     given-names: Yimin
-repository-code: "https://github.com/initial-d/me_fasttext"
-url: "https://arxiv.org/abs/2506.01254"
+repository-code: "https://github.com/initial-d/ngram-store"
+url: "https://github.com/initial-d/ngram-store"
 license: MIT
 keywords:
+  - n-gram memory
+  - vector compression
+  - Engram
   - fastText
   - word embeddings
   - subword embeddings

CONTRIBUTING.md

@@ -1,6 +1,6 @@
 # Contributing
 
-Thanks for taking a look at `me_fasttext`. This is a research prototype, so the
+Thanks for taking a look at NGramStore. This is a research prototype, so the
 most valuable contributions are the ones that make its evidence easier to
 understand, reproduce, or falsify.
 
@@ -18,11 +18,11 @@ Good first contributions include:
 - a public-corpus benchmark report using the
   [evaluator quickstart](docs/evaluator_quickstart.md);
 - an
-  [evaluation question](https://github.com/initial-d/me_fasttext/issues/new?template=evaluation-question.yml)
+  [evaluation question](https://github.com/initial-d/ngram-store/issues/new?template=evaluation-question.yml)
   about build, benchmark setup, corpus preparation, baselines, or result
   interpretation;
 - a
-  [showcase or use-case note](https://github.com/initial-d/me_fasttext/issues/new?template=showcase.yml)
+  [showcase or use-case note](https://github.com/initial-d/ngram-store/issues/new?template=showcase.yml)
   for citations, list entries, comparisons, integrations, or deployment notes;
 - a retrieval, ranking, classification, or OOV-heavy serving result;
 - a negative result showing where the compact `.z` path does not help;
@@ -35,15 +35,15 @@ logs, or credentials. Aggregate metrics and corpus manifests are enough.
 ## Benchmark reports
 
 Use the
-[benchmark result issue template](https://github.com/initial-d/me_fasttext/issues/new?template=benchmark-result.yml)
+[benchmark result issue template](https://github.com/initial-d/ngram-store/issues/new?template=benchmark-result.yml)
 when possible. A useful report should include:
 
-- exact `me_fasttext` commit SHA;
+- exact NGramStore commit SHA;
 - corpus name, language, license status, split, and preprocessing;
 - hardware, OS, compiler, filesystem, and storage type;
 - build, training, export, and benchmark commands;
-- `.bin`, `.vec`, and `.z` artifact sizes;
-- `bench_ftindex` output for compact `.z` serving;
+- full input/table/index/metadata bytes and the exact lossless/approximate mode;
+- dense/raw-mmap controls, native query latency and anonymous/file-backed RSS;
 - one quality metric, or an explicit serving-only caveat;
 - failures, omitted baselines, and anything unfavorable to the method.
 
@@ -55,7 +55,7 @@ for the fuller report shape.
 ## Showcase and visibility notes
 
 Use the
-[showcase issue template](https://github.com/initial-d/me_fasttext/issues/new?template=showcase.yml)
+[showcase issue template](https://github.com/initial-d/ngram-store/issues/new?template=showcase.yml)
 for lighter-weight adoption evidence: citations, public list entries,
 integration notes, talks, blog posts, retrieval experiments, or compact-serving
 deployment notes.
@@ -90,11 +90,7 @@ or platform limitation.
 
 ## Scope and positioning
 
-Keep claims narrow. `me_fasttext` is best framed as a FastText-derived compact
-lexical embedding artifact for memory-sensitive serving, OOV-heavy text, and
-first-stage retrieval or ranking. It is not a replacement for transformer
-embeddings, a vector database, or a full RAG system.
-
+NGramStore is a compression, storage and exact-query backend for learned n-gram vectors. FastText and Engram/TinyEngram are applications with different training and addressing contracts. Numeric/hash queries use the native row engine; structural sharing requires registered string keys. Lossless bytes, approximate distortion, query cost and task quality need separate evidence.
 See [`docs/artifact_card.md`](docs/artifact_card.md) and
 [`docs/citation_guide.md`](docs/citation_guide.md) for wording that is easier
 to cite without overclaiming.