Training - Trelis Studio

Model Training

Base Model

Training Dataset

HuggingFace dataset with 'audio' and 'text' columns. View requirements

Split

Columns: Must have audio and text columns
Audio: Max 30 seconds per sample (Whisper limit)
Text: Max ~1500 characters per sample (~444 tokens, Whisper limit)
Rows: Limited to 100 training / 50 validation rows. Sign in for higher limits.

Validation Dataset (optional)

If empty: 5% of training data (max 50 rows) used for validation

Split

HuggingFace Hub Settings (model + eval results)

Push to HuggingFace Hub

Uploads trained model and evaluation results to Hub

HuggingFace Token

Required for private base models or to push model to Hub. Get token

Output Organization

Output Model Name

Make model private

Weights & Biases Settings (for tracking)

W&B API Key

Track training progress with W&B. Get API key

Advanced Settings

Batch Size

Learning Rate

Epochs

LR Schedule

LoRA Rank (r)

LoRA Alpha

Gradient Accumulation Steps

Use rsLoRA (Re-scaled LoRA)

No job

No training running. Submit a job to see progress here.

Time	Base Model	Dataset	Output	Baseline WER	Final WER	Status
No training jobs yet