mirror of
https://github.com/hanzoai/open-instruct.git
synced 2026-10-30 18:26:32 +00:00
33 lines
940 B
YAML
33 lines
940 B
YAML
model_name_or_path: EleutherAI/pythia-14m
|
|
model_revision: main
|
|
use_flash_attn: true
|
|
tokenizer_name: EleutherAI/pythia-14m
|
|
use_slow_tokenizer: false # olmo models only use fast tokenizers
|
|
# train_file: data/processed/tulu_v2/tulu_v2_data.jsonl
|
|
# dataset_name: allenai/tulu-v2-sft-mixture
|
|
dataset_mixer:
|
|
allenai/tulu-v2-sft-mixture: 100
|
|
# data/processed/tulu_v2/tulu_v2_filtered_data.jsonl: 0.1
|
|
# dataset_mixer:
|
|
# allenai/tulu-v2-sft-mixture: 100000
|
|
# HuggingFaceH4/no_robots: 5000
|
|
max_seq_length: 4096
|
|
max_train_samples: 100
|
|
preprocessing_num_workers: 1
|
|
per_device_train_batch_size: 2
|
|
gradient_accumulation_steps: 8
|
|
learning_rate: 2.0e-05
|
|
lr_scheduler_type: linear
|
|
warmup_ratio: 0.03
|
|
weight_decay: 0.0
|
|
num_train_epochs: 2
|
|
output_dir: output/sft_pythia_14m/
|
|
with_tracking: true
|
|
report_to:
|
|
- wandb
|
|
logging_steps: 1
|
|
add_bos: true # needed for GPTNeoX olmo models
|
|
dataset_mix_dir: null
|
|
checkpointing_steps: 1000
|
|
keep_last_n_checkpoints: 1
|