All notable changes to this project are documented here. Format follows Keep a Changelog; this project adheres to Semantic Versioning.
:download-timeout-ms option for from-pretrained (30000 ms default).batch-count-tokens.token->chars, token->word, char->token, and
word->tokens.ids, tokens, and count-tokens now read native arrays directly, avoiding
full encode-map allocation with unchanged behavior.encode-pretokenized) preserving word IDs.builder escape hatch to pass arbitrary DJL tokenizer options, and from-bpe-files construction from separate vocab.json + merges.txt.encode->ndlist / batch-encode->ndlist for direct DJL inference pipelines, and NDManager-owned tokenizer lifecycle.truncation, padding, max-length, stride, pad-to-multiple-of, effective-config), build-sentence, and native-version.from-file, from-pretrained, from-stream constructors over DJL's
HuggingFaceTokenizer (native Rust tokenizers via JNI).encode returning a Clojure map (:ids :tokens :type-ids :word-ids :attention-mask :special-tokens-mask), with :add-special-tokens? /
:with-overflowing-tokens? options.ids, tokens, count-tokens, decode, and batch-encode helpers.Closeable, so with-open frees the native handle.Can you improve this documentation?Edit on GitHub
cljdoc builds & hosts documentation for Clojure/Script libraries
| Ctrl+k | Jump to recent docs |
| ← | Move to previous article |
| → | Move to next article |
| Ctrl+/ | Jump to the search field |