Simon Suo

- Cluster character by rhyming, and remove generation and limit scope of first character to subset of rhyming ones - Polishing: separate meaning modeling and pronunciation, additional pass to fix...

- https://github.com/google/seq2seq - https://github.com/farizrahman4u/seq2seq - https://www.tensorflow.org/tutorials/seq2seq

### Summary * Introduce concept of `QueryProcessor` and simple implementation of HyDE query processor * Naive integration with `QueryRunner` where we run query processor once * Add HyDE example to...

### Summary * Refactor query interface to use `QueryBundle` instead of raw `str` * This allows users to explicitly use different string(s) for embedding-based queries and non-embedding based queries. ###...

### Summary * Support async predict in `LLMPredictor` * Support naive (per level) async index building in `GPTTreeIndexBuilder` > Use async event loop to process nodes in each level (still...

### Changes to Index construction * main constructor now takes `nodes` instead of `documents`. > We use `Index.from_documents` to continue support parsing documents into nodes under the hood. * Index...

Removing IndexRegistry class, use hard-coded mappings instead since there's not much value for users to customize this mapping.

 ### Conceptual changes * Decouple index structure and data > `V2IndexStruct` is no longer a `BaseDocument`. * We introduce `CompositeIndexStruct` to model a composed graph > This makes...