v0.1.0

✨ Viberary 0.1.0 release!✨

Lots happening, try it out! Viberary.pizza

• Added more documentation
• Fixed an issue that wasn't logging the full query path in the results
• Refined logging for query results
• Working on making deploys automatic for both servers (still in progress)

What's Changed

• Perf by @veekaybee in #81
• Performance improvements and code refactor by @veekaybee in #82
• Additional documentation by @veekaybee in #83
• New docs by @veekaybee in #84
• Refine docs and change logging by @veekaybee in #85
• Second server actions deploy by @veekaybee in #95
• Fix query redirect and add docs by @veekaybee in #98

Full Changelog: v.0.0.9...v.0.1.0

v.0.0.9

What's Changed

• Harden Flask by @veekaybee in #76
• docker-compose changes for gunicorn by @veekaybee in #77
• Fix link to homepage by @veekaybee in #78
• Fix link by @veekaybee in #79
• Load test by @veekaybee in #80

Full Changelog: v.0.0.8...v.0.0.9

viberary 0.0.8

What's Changed

Model has been refactored to use ONNX Runtime at inference time for faster inference.

I switched to an M2 mac for local development which also meant rewriting a lot of Docker workflows.

And as always more unit tests

• Include ONNX model and refactor positional items by @veekaybee in #71
• Add ONNX model features by @veekaybee in #72

Full Changelog: v.0.0.7...v.0.0.8

viberary 0.0.7

What's Changed

Adding query suggestions, more unit tests, and cleaning up the UI

• Adding parquet tests and config file development by @veekaybee in #62
• Add button suggestions by @veekaybee in #68
• Fix buttons by @veekaybee in #69

Full Changelog: v.0.0.6...v.0.0.7

Viberary v.0.0.6

What's Changed

• Makefile driven development and changes in documentation
• Removing duplicate titles from ranked search results
• Starting to abstract away filepaths into config

• Makefile and TOML config by @veekaybee in #57
• Remove duplicates by @veekaybee in #58
• Dedup reranking by @veekaybee in #59
• Fix rerank by @veekaybee in #60

Full Changelog: v.0.0.5...v.0.0.6

v.0.0.5

What's Changed

Quality of life improvements: Link to title to inspect results, a UI prompt to search for vibes, and on the backend more unit testing, Docker compose port, and staging environment link, as well as retrained embeddings

• Reformat with Ruff and fix title/author by @veekaybee in #43
• Adding multiple deploy option templates by @veekaybee in #44
• Harden training data generation class by @veekaybee in #45
• Change flask port to expose staging env by @veekaybee in #47
• Change port for docker compose Flask host by @veekaybee in #48
• Adding links to Viberary results and first indexer test by @veekaybee in #49
• fix unit test by @veekaybee in #50
• Links by @veekaybee in #53
• Fix html by @veekaybee in #54
• Fix html and indexing by @veekaybee in #55
• Fix html by @veekaybee in #56

Full Changelog: v.0.0.4...v.0.0.5

v.0.0.4

What's Changed

The previous SBERT all-MiniLM-L6-v2 was not giving great results, so doing a couple of large changes to improve results of the model:

• Using asymmetric search much more like a search engine would with msmarco-distilroberta-base-v3
• Including more training data: book author and book review from the original goodreads dataset to pull out semantic meaning in the search results
• Reconfiguring Redis indexer to write everything as a hash versus having multiple hash lookups
• Remove Grafana for now - more noise than it's worth
• Including the author in the resultset

• A number of fixes: by @veekaybee in #31
• Use PyTorch CPU only dependencies by @veekaybee in #33
• Retrain embeddings using asymmetric semantic search by @veekaybee in #39
• Fixing Docker Compose for prod by @veekaybee in #41
• Indexing speed by @veekaybee in #42

Full Changelog: v.0.0.3...v.0.0.4

viberary 0.0.3

What's Changed

Creating a hardened Droplet on Digital Ocean to use for testing, rerunning, deleting data, etc, before production. Release includes embeddings Parquet file for inference.

• Refactor datagen by @veekaybee in #17
• Linting by @veekaybee in #18
• Refactoring log delivery by @veekaybee in #24
• Harden Redis to not drop index and data by @veekaybee in #26
• Quick HTML fix by @veekaybee in #27
• Fixing Docker network communication by @veekaybee in #29

Full Changelog: v0.0.2...v.0.0.3

viberary 0.0.2

Breaking changes to API, project still in pre-alpha-alpha

What's Changed

• Enormous BERT Refactor by @veekaybee in #5
• Refactored API into app with /src directory structure instead of exploratory notebooks
• Included learned_embeddings.snappy - trained data from Paperspace training runs 0bc5b01
• Include ranking feature to change cosine similarity to ordinal scores cdfe20b
• Unit tests 3d4f7de
• Dockerized app and can run in docker-compose
• Adding BERT sentence-transformer inference capabilities
• Persistent Redis storage in Docker-compose 8d98db6
• README changes 639ca17

Full Changelog: v0.0.1...v0.0.2

Follow issues here

viberary 0.0.1

Viberary release notes

This is the first "release" of ✨ Viberary✨. I'm releasing this ahead of a major refactor to include the new BERT model and clean up the code into actual OOP + Docker. It includes the artifacts and code needed to run the word2vec model demoed here:

word2vec_viberary.mov

🚧🚧🚧🚧 This code is ridiculously under construction, not production-grade, no unit tests 😿 , no guarantees are made about its quality or the developer's state of mind while constructing it 🚧🚧🚧🚧🚧

To run the models:

1. create a virtualenvironment
2. Download the requirements
3. cd flask-server
4. Download the model artifacts in the release (word2vec and fasttext learned models) into a models directory relative to flask_server
5. cd flask_server and python api.py
6. You "should" see a Flask app on localhost:5000 in your web broswer.

Try inputting some words and switching between models to see what the models return.