# Whisper large-v3-turbo + MOBIUS MMV 12B QAT Voice Formatting Tool — dependencies
# ============================================================
# Installation:
#   pip install -r requirements.txt
#
# To use the GPU (CUDA), install the CUDA build of PyTorch first
# (RTX 50-series Blackwell requires cu128 or later):
#   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
#
# See README.md for preparing Ollama, the evaluated model and the MMV
# harness repository (needed by the local MMV-Format engine and the optional
# cloud path alike; set MMV_REPO if it is not at the default location).
# ============================================================

# Speech transcription
openai-whisper>=20250625

# HTTP client (used by the MMV harness adapters)
requests>=2.31.0

# GPU management (CUDA build of PyTorch recommended; CPU-only also works)
torch>=2.7.0

# Reading MMV release pointers / profiles
pyyaml>=6.0

# ============================================================
# Audio loading calls the ffmpeg binary directly via subprocess from
# whisper, so the Python package ffmpeg-python is NOT required.
# Only the ffmpeg binary itself must be installed:
#   Ubuntu/Debian: sudo apt install ffmpeg
#   Mac:           brew install ffmpeg
#   Windows:       download from https://ffmpeg.org/download.html and add to PATH
# ============================================================