version bump

Feat/indexing faissless (#173)

* fix: fix searcher always being reloaded

* feat: implement torch kmeans

* chore: lower cutoff

* chore: move warning

* chore: higher kmeans batch size

* chore: argument support

* chore: restore all default behaviour when `use_faiss` is True after having been false

* chore: lint

* chore: print exception if one occurs when using pytorch indexing

* chore: make _original_train_kmeans robust to subsequent calls

* nit: comment

feat: rework kmeans to be closer to FAISS

chore: store kmeans functions as class attributes

fix: method assignment

chore: more memory efficient

lint

chore: lower bsize, resultd unaffected

feat: better batching, slower max doc count

chore: batch size safe for 8gb GPUs

chore: more elaborate warning

chore: use external lib to support minibatching, revert to homebrew later

* poetry lock

* lint

* chore: better batch size

* 0.0.8 dependency prep

version bump

Chore/lower bs for long docs (#134)

* chore: auto-adjust batch size for longer docs

* chore: move function to more appropriate file

* fix bsize calc

fix: poetry lock

Fix: dynamically increase query params for higher `k` values (#131)

* fix: return enough results if k > ncells*32

* fix: increase both ndocs and ncells to match k

* chore: prepare release

* linting

* chore: saner ncells for larger datasets

* linting

chore: format & minor fixes

chore: ✏️ updating hyperlink (#88)

Co-authored-by: Almonok <[email protected]>

readme: download badge

chore: Fixes, better param handling (#77)

* fix: miner fixes, encoding length

* chore: pre-release