Repository navigation
Reposition as NGramStore with compatible engine APIs and working quickstarts - #37
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The repository's primary identity is now NGramStore: a compression, storage and exact-query engine for learned n-gram vectors. FastText and Engram/TinyEngram are application examples of the shared backend. The README now gives one engine entry, application routes and measured limits, while historical FastText usage moves to a compatibility guide.
Add the
ngram_storeimport andngram-storecommand without removingmeft_memory, themeft-memorydistribution, existingmeft-*commands, C ABI symbols or file-format identifiers. The new dispatcher packs/inspects/queries native stores, verifies every row before publication, supports registered-key structural planning, and routes to existing FastText/TinyEngram converters. Numeric IDs, multiplicity and source bits remain intact; joint sharing remains explicitly approximate.The old plain-text training quickstart did not match the custom FastText reader. A preparation command now emits within-line positional/source pairs, with a tiny included text sample and an actual Linux prepare/train/export/query regression. This is integration correctness, not a trained compression benchmark. Canonical repository URLs and software metadata target
initial-d/ngram-store; the original paper title/preferred citation and dated raw evidence remain intact. The authorized GitHub repository rename follows this PR's checks and merge.Validation: portable engine tests preserve signed-zero/NaN payloads and original IDs; registered-key tests exercise actual DATrie compaction; installed wheels verify new and historical imports plus the new CLI outside the checkout without Torch; Torch opens the same packed artifact; the existing FastText/Engram/Qwen and sanitizer checks remain in CI. Local targeted tests and isolated macOS wheel checks pass. No GPU workload is used.