wandb/run-20250527_212216-prr8gv5v/files/config.yaml

_wandb:
    value:
        cli_version: 0.19.11
        m: []
        python_version: 3.11.11
        t:
            "1":
                - 1
                - 5
                - 11
                - 41
                - 49
                - 51
                - 53
                - 55
                - 71
                - 105
            "2":
                - 1
                - 5
                - 11
                - 41
                - 49
                - 51
                - 53
                - 55
                - 71
                - 105
            "3":
                - 2
                - 13
                - 16
                - 23
                - 55
                - 61
            "4": 3.11.11
            "5": 0.19.11
            "6": 4.52.1
            "8":
                - 5
            "12": 0.19.11
            "13": linux-x86_64
bf16:
    value: true
deepscale:
    value: false
deepscale_config:
    value: null
deepspeed:
    value: false
deepspeed_config:
    value: null
device:
    value: cuda:0
epochs:
    value: 1
eval_datasets:
    value: null
eval_interval:
    value: 1000000
eval_split_ratio:
    value: null
eval_strategy:
    value: epoch
fp16:
    value: false
global_rank:
    value: 0
gradient_accumulation_steps:
    value: 8
gradient_checkpointing:
    value: true
local_rank:
    value: 0
log_dir:
    value: /aifs4su/hansirui_1st/jiayi/setting3-imdb/tinyllama-3T/tinyllama-3T-s3-Q1-2000-Q2-1000
log_project:
    value: Inverse_Alignment_IMDb
log_run_name:
    value: imdb-tinyllama-3T-s3-Q1-2000-Q2-1000
log_type:
    value: wandb
lr:
    value: 1e-05
lr_scheduler_type:
    value: CONSTANT
lr_warmup_ratio:
    value: 0
max_length:
    value: 512
model_name_or_path:
    value: /aifs4su/hansirui_1st/jiayi/setting3-imdb/tinyllama-3T/tinyllama-3T-s3-Q1-2000
need_eval:
    value: false
num_update_steps_per_epoch:
    value: 16
offload:
    value: none
output_dir:
    value: /aifs4su/hansirui_1st/jiayi/setting3-imdb/tinyllama-3T/tinyllama-3T-s3-Q1-2000-Q2-1000
per_device_eval_batch_size:
    value: 4
per_device_train_batch_size:
    value: 1
save_16bit:
    value: true
save_interval:
    value: 1000000
seed:
    value: 42
tf32:
    value: true
total_training_steps:
    value: 16
train_datasets:
    value:
        - - inverse-json
          - path: /home/hansirui_1st/jiayi/resist/imdb_data/train/neg/1000/train.json
            proportion: 1
trust_remote_code:
    value: true
weight_decay:
    value: 0
zero_stage:
    value: 3