Skip to content
aiea-labPublic

About

Custom LLM for the AIEA Lab (for language in hawaiian: Ōlelo)

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

31 Commits

Folders and files

Repository files navigation

ʻŌlelo

ʻŌlelo (Hawaiian: language, speech, expression) is a custom large language model developed by the AIEA Lab to support research, teaching, and experimentation in explainable and neurosymbolic AI.

The goal of ʻŌlelo is not just to generate a custom LLM for the AIEA lab to support structured reasoning, transparency, and responsible use of language models in large-group, academic settings.


🌺 About the Name

In ʻŌlelo Hawaiʻi, ʻōlelo refers to language, speech, and meaningful expression.
The name reflects our view of language models as systems that express knowledge, rather than opaque black boxes that merely predict tokens.


🎯 Project Goals

ʻŌlelo is designed to:

  • Listen to meetings and produce meeting notes
  • Be a custom tutor and helper for the lab (discord bot)
  • Create workflows (automatically create Notion meeting notes, etc.)

Eventually, the project will:

  • Support research workflows in explainable AI (XAI) and neurosymbolic methods
  • Serve as a teaching and mentoring aid for students
  • Enable controlled experimentation with reasoning, evaluation, and grounding
  • Provide a platform for studying what models know, don’t know, and how they explain it

High level overview

  1. Wait for Zoom Webhook to fire. Zoom will send events, when a meeting is completed.
  2. After an event is sent, the pipeline will capture all audio files from the specific meeting session then transcribe (Whisper) and diarize (Pyannote) them.
  3. The diarizations will get matched with transcriptions based on timestamps.
  4. Gemma (in-house LM) recieves the merged transcripts and diarizations and gives a brief summary of the meeting.

This project is intended primarily for internal AIEA lab and academic use.

Transcribing Video/Audio Files with Whisper

Whisper is an open source transcription model created by OpenAI.

Name Parameters English-only model Multilingual model Required VRAM Relative Speed
tiny 39 M tiny.en tiny ~1 GB ~10x
base 74 M base.en base ~1 GB ~7x
small 244 M small.en small ~2 GB ~4x
medium 769 M medium.en tiny ~5 GB ~10x
large 1550 M N/A large ~10 GB 1x
turbo 809 M N/A turbo ~6 GB ~8x

Accepted File Types

Audio: 3GA, 8SVX, AAC, AC3, AIF, AIFF, ALAC, AMR, APE, AU, DSS, FLAC, M4A, M4B, M4P, M4R, MOGG, MP2, MP3, MPGA, OGG, OGA, OPUS, QCP, TTA, VOC, WAV, WMA
Video: FLV, M2TS, M4V, MKV, MOV, MP4, MTS, MXF, TS, WEBM
(Up to 5GB per file)

TO USE:

  1. Clone this git repo:
git clone https://github.com/aiea-lab/olelo.git 
cd olelo
  1. Install openai-whisper and its dependencies:
pip install openai-whisper moviepy
  1. Place files in olelo/transcription_pipeline/files_to_transcribe:
mkdir ./transcription_pipeline/files_to_transcribe
mv <path_to_your_files> ./transcription_pipeline/files_to_transcribe
  1. Make sure you are inside the olelo directory and paste this script into your terminal (replace <modeltype> with one of the models above):
python ./transcription_pipeline/whisper_transcribe.py <model_type>
  1. View your results in transcription_pipeline/transcriptions.txt

About

Custom LLM for the AIEA Lab (for language in hawaiian: Ōlelo)

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages