Skip to content

armgit / Soup

Public Imported from GitHub
0

Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.

771 B
Raw Blame
# Soup template: Audio / Speech
# Fine-tune an audio-language model for speech understanding
#
# Supported models: Qwen2-Audio, Whisper (via transformers)
#
# Data format (JSONL):
#   {"audio": "path/to/audio.wav", "messages": [
#     {"role": "user", "content": "Transcribe."},
#     {"role": "assistant", "content": "Hello world."}]}

base: Qwen/Qwen2-Audio-7B-Instruct
task: sft
modality: audio
backend: transformers  # The audio SFT path does not use Unsloth.

data:
  train: ./data/audio_train.jsonl
  format: audio
  audio_dir: ./data/audio
  val_split: 0.1
  max_length: 2048

training:
  epochs: 3
  lr: 1e-5
  batch_size: auto
  gradient_accumulation_steps: 8
  lora:
    r: 64
    alpha: 16
    target_modules: auto
  quantization: 4bit

output: ./output_audio