mirror of
https://github.com/hanzoai/open-instruct.git
synced 2026-11-03 02:13:07 +00:00
Yizhong WangandGitHub
c86c6a098c
* The final v3.4 mixture is a copy from v3.4.23, performing best across these configurations.
24 lines
721 B
YAML
24 lines
721 B
YAML
model_name_or_path: meta-llama/Meta-Llama-3-8B
|
|
model_revision: main
|
|
use_flash_attn: true
|
|
tokenizer_name: meta-llama/Meta-Llama-3-8B
|
|
use_slow_tokenizer: true
|
|
dataset_mixer:
|
|
HuggingFaceH4/no_robots: 9500 # all
|
|
max_seq_length: 4096 # Note, reduced from 8192 to fit on one GPU with DeepSpeed Stage3
|
|
preprocessing_num_workers: 128
|
|
per_device_train_batch_size: 1 # note, this is set up for 8 GPUs
|
|
gradient_accumulation_steps: 4 # effective batch size 128 with 4 nodes
|
|
learning_rate: 5.0e-06 # best LR so far
|
|
lr_scheduler_type: linear
|
|
warmup_ratio: 0.03
|
|
weight_decay: 0.0
|
|
num_train_epochs: 3
|
|
output_dir: /output/
|
|
with_tracking: true
|
|
report_to:
|
|
- wandb
|
|
logging_steps: 1
|
|
checkpointing_steps: epoch
|
|
dataset_mix_dir: /output/
|