andreaskoepf/

pythia-1.4b-gpt4all-pretrain

pretrain:
  weight_decay: 0.01
  use_custom_sampler: true
  sort_by_length: false
  datasets:
    - joke
    - webgpt:
        val_split: 0.1
    - gpt4all:
        val_split: 0.01
    - alpaca:
        val_split: 0.025
    - code_alpaca:
        val_split: 0.05
    - minimath
    - humaneval_mbpp_codegen_qa
    - humaneval_mbpp_testgen_qa
    - grade_school_math_instructions
    - recipes
    - cmu_wiki_qa
    - oa_wiki_qa_bart_10000row
    - prosocial_dialogue:
        fraction: 0.1
    - explain_prosocial:
        fraction: 0.05

pythia-1.4b-pretrain:
  dtype: fp16
  learning_rate: 6e-6
  model_name: EleutherAI/pythia-1.4b-deduped
  deepspeed_config: configs/zero_config_pretrain.json
  weight_decay: 0.0
  max_length: 2048
  use_flash_attention: true
  warmup_steps: 50
  gradient_checkpointing: false
  gradient_accumulation_steps: 1
  per_device_train_batch_size: 16
  per_device_eval_batch_size: 16
  num_train_epochs: 2 
  save_total_limit: 2