Files
pythia410m-sft-tldr/code/configs/sft_pythia1b_fp16_trainall_3epochs_4V100.yml
ModelHub XC 4bb3a5553f 初始化项目,由ModelHub XC社区提供模型
Model: mnoukhov/pythia410m-sft-tldr
Source: Original Platform
2026-05-30 14:21:21 +08:00

19 lines
404 B
YAML

model_name: EleutherAI/pythia-1b-deduped
log_with: "wandb"
learning_rate: 1e-5
lr_scheduler_type: cosine
num_warmup_steps: 100
weight_decay: 0.05
load_in_8bit: False
use_peft: False
fp16: True
bf16: False
streaming: False
gradient_checkpointing: False
gradient_accumulation_steps: 4
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
train_completions: False
num_train_epochs: 3
packing: False