Transcribe whole meetings with speaker attribution: open-weights model, benchmark meetings, replication tooling.
AI & ML interests
Fine-tuning, Training and Deploying Large Language, Vision and Audio models.
Recent Activity
Team members 2 private
Mimi, SNAC and BPE Tokenisations of LibriTTS (Parler's clean split)
Pre-trained Piper TTS voices in ONNX format for fast CPU inference. Fine-tune on your own data at studio.trelis.com.
A dataset of Python Programs solving ARC AGI Tasks
Versions of the OpenAssist Dataset for chat fine-tuning different models. See https://youtu.be/71x8EMrB0Gc for a full video run through of methods.
Tara frontier Hindi/Hinglish transcription model and the evaluation datasets behind its benchmark numbers. Eval code: github.com/TrelisResearch/tara
Datasets for Medical ASR / Transcription
Fine-tuned Whisper Models for cleaner transcription to GB or US English
-
Trelis/transcribe-to-en_GB-v1
Viewer • Updated • 30 • 16 • 1 -
Trelis/transcribe-to-en_US-v1
Viewer • Updated • 30 • 18 -
Trelis/transcribe-en_gb-spelling-v1-tiny
Automatic Speech Recognition • 37.8M • Updated • 22 -
Trelis/transcribe-en_gb-spelling-v1-tiny-ctranslate2
Automatic Speech Recognition • Updated • 5
TRM Models
Models fine-tuned for function-calling
-
Trelis/Meta-Llama-3-8B-Instruct-function-calling
Text Generation • 8B • Updated • 54 • • 45 -
Trelis/openchat_3.5-function-calling-v3
Text Generation • 7B • Updated • 21 -
Trelis/Llama-2-7b-chat-hf-function-calling-v3
Text Generation • 7B • Updated • 688 • • 41 -
Trelis/Yi-34B-200K-Llamafied-chat-SFT-function-calling-v3
Text Generation • 34B • Updated • 5
Transcribe whole meetings with speaker attribution: open-weights model, benchmark meetings, replication tooling.
Tara frontier Hindi/Hinglish transcription model and the evaluation datasets behind its benchmark numbers. Eval code: github.com/TrelisResearch/tara
Mimi, SNAC and BPE Tokenisations of LibriTTS (Parler's clean split)
Datasets for Medical ASR / Transcription
Pre-trained Piper TTS voices in ONNX format for fast CPU inference. Fine-tune on your own data at studio.trelis.com.
Fine-tuned Whisper Models for cleaner transcription to GB or US English
-
Trelis/transcribe-to-en_GB-v1
Viewer • Updated • 30 • 16 • 1 -
Trelis/transcribe-to-en_US-v1
Viewer • Updated • 30 • 18 -
Trelis/transcribe-en_gb-spelling-v1-tiny
Automatic Speech Recognition • 37.8M • Updated • 22 -
Trelis/transcribe-en_gb-spelling-v1-tiny-ctranslate2
Automatic Speech Recognition • Updated • 5
A dataset of Python Programs solving ARC AGI Tasks
TRM Models
Versions of the OpenAssist Dataset for chat fine-tuning different models. See https://youtu.be/71x8EMrB0Gc for a full video run through of methods.
Models fine-tuned for function-calling
-
Trelis/Meta-Llama-3-8B-Instruct-function-calling
Text Generation • 8B • Updated • 54 • • 45 -
Trelis/openchat_3.5-function-calling-v3
Text Generation • 7B • Updated • 21 -
Trelis/Llama-2-7b-chat-hf-function-calling-v3
Text Generation • 7B • Updated • 688 • • 41 -
Trelis/Yi-34B-200K-Llamafied-chat-SFT-function-calling-v3
Text Generation • 34B • Updated • 5