v2.2.1 - Whisper easier

Since the transcriber is set to "auto" to select best available of gpu or cpu acceleration. Therefore, I added 8 threads default (when using cpu) instead of 4, which is the default from faster-whisper library when not otherwise specified.

Added instructions specific to BGE embedding models, which was accidentally overlooked previously, which will greatly increase the effectiveness of these models (now rivaling or possibly exceeding the instructor embeddings models' quality).

v2.2 - Whisper quants

Simplified the process to download and use different Whisper/Ctranslate2 models automatically.

Revised instructions.

v2.1 - transcribe question!

Added audio transcription for your question to the text box!

Refactored multiple scripts for further expansion and modularity.

v2.0.1 - refactor

Refactored to use initialize.py in order to shorten gui.py somewhat.

Commented out a portion of the html file displayed in the tab. Didn't want to offend the hard work of the ggml/gguf/gptq people unless/until I can more thoroughly test ctranslate2 and their implementations.

Stopped the left pane from hogging the GUI's space, which had made the question and answer boxes too small.

v2.0 - PySide6!

1 - Major refactoring of all scripts.

2 - Upgrade GUI to use PySide6.

3 - Revamp GUI to start using tabs.

4 - CHECK OUT the "Instructions" tab for what's to come.

v1.7 - Important

Removed releases between 1.4.2 and 1.6 because the newer version of chromadb was causing problems; somehow has a limit of ~5600 chunks, which prevents ingesting very large PDF document. The only thing I could think of is that the newer ChromaDB relies on Sqlite3 while the older version relies on DuckDB + Parquet/Clickhouse...So rather than figure out how to make the newer ChromaDB work, reverted, works fine.

This release includes all prior improvements from releases I deleted, except the newer ChromaDB version. Thus, new GUI, new layout, refactoring of scripts in preparation for expansion, numerous additional embedding models, etc.

v1.4.3 - models galore!

ADDED a bunch of new embedding models to choose from!

Renamed scripts and functions to prepare for expansion.

Removed a bug that prevented it from working on Macs and AMD Gpus (related to displaying metrics).

If anyone has any bugs please let me know. I will implement a custom feature you request (within reason) if you report a bug to me that make the program work better!

v1.4.2 - more metrics!

Added cpu and ram usage and percentage metrics.

Refactored code to prepare for expansion.

Roadmap:
--Introduce quantized embedding models for even faster increase and lower resource requirement.
--Add options when creating the database and interacting with the LLM; for example, controlling the chunk size or the number of results or the length of the results...to make sure it fits within the LLM's context window.
--Add a calculator that displays the total tokens of a user's prompt + the context returned to a user can see if it fits within the LLM's context window.
--Add "poor man's vector database" search based on my other repo. Make this an option within this program.
--Add some color and other stuff to improve ease on eyes and appearance of the GUI.
--Remove the table on the left side that shows embedding models. Add a "help" or user's manual with various tables this information (only when requested), among other information like primers on how a vector database works, how to get the most out of it, nuances of the various models, and other helpful stuff.

v1.4.1 - cuda/vram/multiprocessing/threading

Properly implemented multithreading/processing to make sure the CUDA/VRAM usage (and the GUI in general) doesn't freeze when creating the vector database nor when querying the database.

Updated pro tip to reflect reliable comments on Discord regarding larger LLMs being helpful for especially technical jargon.

v1.4 - BREAKING changes

Significantly revised the code, created new scripts, started using a configuration yaml file, reducing the "global variables," etc.

After much struggling, added a GPU and VRAM usage metrics at the bottom of the GUI so you can see when running! However, the GUI still hangs periodically, which prevents it from updating when you need it most. This will be fixed in a patch in the next day or so.

Comments are welcome. Collaboration is appreciated.