# FancyAI English memory package

Optional local memory models for FancyAI. Installing this package does not enable character memory automatically.

- **Summarization:** `l3utterfly/bart-large-ltm`, revision `9e446006721f6efeb5bb3760feec387d469d2746`, converted with CTranslate2 4.8.2 to int8.
- **Embeddings:** `Xenova/bge-m3`, revision `4de13258303883538bd53b696b452bf8099f0858`, quantized ONNX model and tokenizer. BGE-M3 uses the first-token vector with L2 normalization.
- Both upstream repositories declare MIT licensing. See their model cards: https://huggingface.co/l3utterfly/bart-large-ltm and https://huggingface.co/Xenova/bge-m3.

This package targets English memory summaries. It is not represented as an exact copy of Layla's installed model package. Summaries can be inaccurate and remain editable in FancyAI.

Download: [fancy-memory-english-v1.zip](./fancy-memory-english-v1.zip)

SHA-256: `7c103dea90f0d92c6af70734b8c84e42a7012fe9cbd78a4e022f6b9649fd95a3`

Extracted size: 1,006,221,253 bytes. Contents: BART model/config/vocabulary/tokenizer, BGE model/tokenizer, and provenance notice.