README.md

metadata

language:
  - en
tags:
  - NLP
license: mit
datasets:
  - TristanBehrens/bach_garland_2024-100K
base_model: None

bach_garland_mambaplus - An xLSTM Model

Trained with Helibrunna by Dr. Tristan Behrens.

Configuration

training:
  model_name: bach_garland_mambaplus
  batch_size: 8
  lr: 0.001
  lr_warmup_steps: 5000
  lr_decay_until_steps: 50000
  lr_decay_factor: 0.001
  weight_decay: 0.1
  amp_precision: bfloat16
  weight_precision: float32
  enable_mixed_precision: true
  num_epochs: 8
  output_dir: output/bach_garland_mambaplus
  save_every_step: 500
  log_every_step: 10
  wandb_project: bach_garland
  torch_compile: false
model:
  type: mamba
  d_model: 128
  n_layers: 8
  context_length: 4096
  vocab_size: 178
dataset:
  hugging_face_id: TristanBehrens/bach_garland_2024-100K
tokenizer:
  type: whitespace
  fill_token: '[EOS]'