_n_gpu: 1
accelerator_config:
  dispatch_batches: null
  even_batches: true
  gradient_accumulation_kwargs: null
  non_blocking: false
  split_batches: false
  use_configured_state: false
  use_seedable_sampler: true
adafactor: false
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1.0e-08
auto_find_batch_size: false
average_tokens_across_devices: true
batch_eval_metrics: false
bf16: true
bf16_full_eval: false
data_seed: null
dataloader_drop_last: false
dataloader_num_workers: 0
dataloader_persistent_workers: false
dataloader_pin_memory: true
dataloader_prefetch_factor: null
ddp_backend: null
ddp_broadcast_buffers: null
ddp_bucket_cap_mb: null
ddp_find_unused_parameters: null
ddp_timeout: 1800
debug: []
deepspeed: ./ds_configs/zero2.json
disable_tqdm: false
do_eval: true
do_predict: false
do_train: false
eval_accumulation_steps: null
eval_delay: 0
eval_do_concat_batches: true
eval_on_start: false
eval_steps: null
eval_strategy: !!python/object/apply:transformers.trainer_utils.IntervalStrategy
- epoch
eval_use_gather_object: false
fp16: false
fp16_backend: auto
fp16_full_eval: false
fp16_opt_level: O1
freeze_backbone: false
freeze_only_attn: false
fsdp: []
fsdp_config:
  min_num_params: 0
  xla: false
  xla_fsdp_grad_ckpt: false
  xla_fsdp_v2: false
fsdp_min_num_params: 0
fsdp_transformer_layer_cls_to_wrap: null
full_determinism: false
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_checkpointing_kwargs: null
greater_is_better: null
group_by_length: false
half_precision_backend: auto
hub_always_push: false
hub_model_id: null
hub_private_repo: null
hub_revision: null
hub_strategy: !!python/object/apply:transformers.trainer_utils.HubStrategy
- every_save
hub_token: null
ignore_data_skip: false
include_for_metrics: []
include_inputs_for_metrics: false
include_num_input_tokens_seen: 'no'
include_tokens_per_second: false
jit_mode_eval: false
label_names: null
label_smoothing_factor: 0.0
learning_rate: 0.0001
length_column_name: length
liger_kernel_config: null
load_best_model_at_end: false
local_rank: 7
log_level: passive
log_level_replica: warning
log_on_each_node: true
logging_dir: ./checkpoints/qwen3-vl-2b_visdoc_baseline/runs/May04_03-23-10_crossnode2
logging_first_step: false
logging_nan_inf_filter: true
logging_steps: 1.0
logging_strategy: !!python/object/apply:transformers.trainer_utils.IntervalStrategy
- steps
lr_scheduler_kwargs: {}
lr_scheduler_type: !!python/object/apply:transformers.trainer_utils.SchedulerType
- cosine
max_grad_norm: 1.0
max_steps: -1
metric_for_best_model: null
model_max_length: 1024
mp_parameters: ''
neftune_noise_alpha: null
no_cuda: false
num_train_epochs: 4.0
optim: !!python/object/apply:transformers.training_args.OptimizerNames
- adamw_torch_fused
optim_args: null
optim_target_modules: null
output_dir: ./checkpoints/qwen3-vl-2b_visdoc_baseline
overwrite_output_dir: false
parallelism_config: null
past_index: -1
per_device_eval_batch_size: 64
per_device_train_batch_size: 64
per_gpu_eval_batch_size: null
per_gpu_train_batch_size: null
prediction_loss_only: false
project: huggingface
push_to_hub: false
push_to_hub_model_id: null
push_to_hub_organization: null
push_to_hub_token: null
ray_scope: last
remove_unused_columns: false
replace_embed_layers: false
report_to:
- wandb
restore_callback_states_from_checkpoint: false
resume_from_checkpoint: null
run_name: qwen3-vl-2b_visdoc_baseline
save_on_each_node: false
save_only_model: false
save_safetensors: true
save_steps: 500
save_strategy: !!python/object/apply:transformers.trainer_utils.SaveStrategy
- epoch
save_total_limit: 1
seed: 42
skip_memory_metrics: true
starting_layer: null
tf32: true
torch_compile: false
torch_compile_backend: null
torch_compile_mode: null
torch_empty_cache_steps: null
torchdynamo: null
tpu_metrics_debug: false
tpu_num_cores: null
trackio_space_id: trackio
train_vision_encoder: false
train_vision_projector: false
use_cpu: false
use_flash_attn: false
use_hard_negatives: false
use_legacy_prediction_loop: false
use_liger_kernel: false
use_mps_device: false
use_steer_layers: false
vision_projector_lr: null
warmup_ratio: 0.03
warmup_steps: 0
weight_decay: 0.0