Fritter Iterations

Fritter Iterations One row per released iteration of Fritter, a fine-tune of Parakeet TDT 0.6B v2, and its streaming twin, with the data that went into it. Newest last. Updated after every release. Column Meaning model, version, label, run Model, its version, the release label and the training run id date Release (or training) date base_model The checkpoint family the line was first fine-tuned from method incremental (continued from the previous weights) or from… See the full description on the dataset page: https://huggingface.co/datasets/danielrosehill/Fritter-Iterations.

About This Project

Fritter-Iterations is a comprehensive, tabular dataset hosted on Hugging Face that documents every released iteration of the Fritter speech-to-text model. It provides a historical record of training runs, including the base model checkpoints, specific release labels, and the incremental methodology used to refine the model's performance over time.

The project utilizes structured formats to detail the lineage of both the primary Fritter model and its streaming variant. By mapping training metadata such as run IDs and release dates, this resource serves as a transparent changelog and reference for researchers and developers interested in the iterative fine-tuning process of the Parakeet TDT architecture.

Project Details

Tags

license:cc-by-4.0size_categories:n<1kformat:jsonmodality:tabularmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:usasrmodel-trainingchangelog

Explore More Projects