DocChatAgent: new re-rankers: * rerank_with_diversity to increase diversity among top chunks * rerank_to_periphery to have best chunks at periphery (mitigates lost-in-middle effect)
0.1.106
Class method: langroid.language_models.base.LanguageModel.usage_cost_summary(): returns summary of token usage, cost aggregated by LLM model
0.1.105
Globally track, report token usage & cost by model-name. E.g. see: * examples/docqa/doc-chat-multi-llm.py * tests/main/test_token_usage.py
0.1.104
DocChatAgent edge case in `add_context_window`: handle absence of window_ids list in chunks
0.1.103
Minor fix in Agent tool handler: re-raise error since we know it's not a Pydantic failure
0.1.102
DocChatAgent: set n_neighbor_chunks > 0 to retrieve window around matching chunk