model_name: allenai/Olmo-3-1025-7B
forget_path: unlearning-cleanslate/fsid-curated-olmo-7b:forget:baseline
retain_path: unlearning-cleanslate/fsid-curated-olmo-7b:retain:baseline
output_dir: /runs
objective: simnpo
num_epochs: 2
batch_size: 8
gradient_accum_steps: 4
lr: 5.0e-05
weight_decay: 0.0
warmup_steps: 10
max_grad_norm: 1.0
forget_weight: 5.0
retain_weight: 1.0
retain_loss_type: nll
npo_beta: 2.0
simnpo_delta: 0.25
rmu_steering_coeff: 2.0
rmu_module_regex: model\.layers\.7
rmu_control_vec_seed: 0
undial_beta: 10.0
save_every: 200
fsdp_sharding_strategy: full_shard
dtype: bfloat16
seed: 42
num_workers: 0
log_every: 10
gradient_checkpointing: false
torch_compile: false
push_to_hub: true
hub_repo_id: unlearning-cleanslate/olmo-3-7b-simnpo-baseline
hub_private: false
optim: adamw_torch
_run_name: simnpo_Olmo-3-1025-7B_20260428_231931
_run_dir: /runs/simnpo_Olmo-3-1025-7B_20260428_231931