crispembed 0.16.1

Lightweight embedding + math OCR inference via ggml. Dense, sparse, ColBERT, cross-encoder reranking, and DeiT+TrOCR on-device math equation recognition.

0.16.1

• Native libraries are now fetched from the CI release on install (no manual placement); iOS/Android build only the crispembed library target.
• Server gains POST /rerank (cross-encoder) and POST /sparse (SPLADE/BGE-M3) routes; /health reports reranker/sparse/colbert capabilities.
• First pub.dev release carrying the v0.16.x native library (0.16.0 shipped as GitHub release binaries only).

0.16.0

• Community & official GGUF interoperability: load third-party llama.cpp / Ollama GGUFs across BPE/WordPiece/SentencePiece tokenizers (ModernBERT, EmbeddingGemma, LFM2.5-Embedding, Qwen3-Embedding, granite-embedding, GTE-v1.5, nomic-v2-moe).
• SPLADE-v3 sparse retrieval model; escaping-aware JSON server input; community-GGUF regression matrix with HF ground-truth parity.

0.15.1

• Docs/quality: add an example, and enable the flow-control brace lint with the corresponding fixes (pub.dev static-analysis + example points).

0.15.0

• Initial pub.dev release of the CrispEmbed Flutter/Dart FFI package.
• Exposes bindings for embeddings, reranking, OCR/OMR, layout detection, scan cleanup and document-processing helpers.
• Supports desktop, mobile and FFI-based native library loading.