88 lines (68 loc) · 2.08 KB

Change Log

0.0.21

Added

Enabled chat_history to be passed to gen_response function.
Added FastAPI support

0.0.20

Fixed

Fixed bug with memory limit

0.0.19

Fixed

Fixed problem with context prompt for non-RAG models

0.0.18

Added

Added condense_plus_context option for chat mode in gen_response, which is better at answering follow-up questions with RAG.
Added ability to tweak context_prompt in gen_response

0.0.17

Fixed

Fixed issue with roles in pg_dump and pg_restore

0.0.16

Fixed

Bug fix in pg_dump and pg_restore functions

0.0.15

Fixed

Bug fix in pgdump

0.0.14

Added

Added dockerfiles and instructions
Added functions to dump and restore vector databases

0.0.13

Fixed

memory_limit in chat mode would previously overload the context window and stop working, fixed now.
context_window moved to be only at LLM instantiation, removed from other locations in gen_response which didn't work.

0.0.12

Added

split the LLM from the vector DB/chat engine model, meaning now you can have multiple separate model objects use the same LLM. Temperature, context window, max new tokens, system prompt, etc. can also all be changed at inference time via the model.gen_response() function.

0.0.11

Added

added automatic handling of CSV data files by converting them to chunked markdown tables

0.0.10

Added

added streaming option to .gen_response

0.0.9

Added

'mps' device

0.0.8

Fixed

separated dropping database and dropping table

0.0.7

Fixed

persistent vector databases

0.0.6

Added

ability to have persistent sessions with a chat engine

0.0.5

Added

ability to change chunk_overlap and paragraph_separator parameters in SentenceSplitter

0.0.4

Added

ability to close vector database connection

0.0.3

Added

two additional methods for converting CSVs to LLM-readable files

0.0.2

Added

ability to convert tabular CSVs to LLM-readable text files

0.0.1

Added

initial release