Fleet-2 Copyright 2026 Lucasgitguy-co. Licensed under the Apache License, Version 2.0 (see LICENSE). This model is a derivative of: - Whistle by Cactus Compute (Apache-2.0): architecture, base weights, runtime. - Sky-Kim/whistle-finetuned-ko (Apache-2.0): Korean vocabulary, language tags, LoRA adapter. - Lucasgitguy-co/Fleet-1 (Apache-2.0): adapter lineage. Training targets were produced by openai/whisper-large-v3-turbo (MIT License, Copyright (c) 2022 OpenAI), run through the mlx-community/whisper-large-v3-turbo conversion. No Whisper weights are distributed here. Training and evaluation audio (not redistributed): - VoxPopuli, Wang et al. (2021), CC0 1.0. - Zeroth-Korean, Atlas Guide Inc. and Gridspace Inc., OpenSLR 40, CC BY 4.0. Transcripts were replaced by teacher output. - FLEURS, Conneau et al. (2022), CC BY 4.0. Audio converted to 16 kHz mono PCM and subsetted.