Add files using upload-large-folder tool

rl_lladav_video_reversal_rl/ckpt/metadata.json

@@ -0,0 +1,3 @@
+{
+  "save_time": "2026-03-27 13:35:06"
+}

rl_lladav_video_reversal_rl/config.yaml

@@ -0,0 +1,102 @@
+wandb:
+  entity: null
+  resume: auto
+  run_id: r7a1v7of
+system:
+  HTTP_PROXY: null
+  HF_HOME: null
+  env_name: /data8/runqing/miniconda/envs/dllm-rl
+  envs_dir: /data8/runqing/miniconda/envs/dllm-rl
+  base_dir: /data4/runqing/dLLM-RL
+experiment:
+  project: rl_lladav_video_reversal_rl
+  function: train
+  start_from_scratch: true
+  total_step: 30
+  save_every: 10
+  eval_every: 5
+  current_epoch: 30
+  deepspeed_file: 1_node_8_gpus_deepspeed_zero2_rl
+  num_node: 1
+  node_index: 0
+  logging_dir: rl_lladav_video_reversal_rl/logs
+model:
+  pretrained_model: /data4/runqing/dLLM-RL/LLaDA-V
+  optimized_name: optimized
+  model_base: lladav
+  vq_model_path: /data4/runqing/hf/magvitv2
+  image_resolution: 512
+dataset:
+  train_dataset: video_reversal_rl_train
+  data_type: video_reverse
+  optimization_data: rl_data
+rollout:
+  num_task_per_step: 128
+  num_response_per_task: 8
+  temperature: 0.8
+  steps: 128
+  max_gen_length: 128
+  batch_size: 1
+  remasking_strategy: low_confidence_dynamic
+  target: confidence
+  dynamic_threshold: 0.9
+  block_size: 32
+  further_horizon: 128
+  use_cache: true
+reward:
+  answer_must_in_box: true
+  strict_len_check: false
+execute:
+  num_chunk: 128
+training:
+  gradient_accumulation_steps: 2
+  batch_size_lm: 1
+  mixed_precision: bf16
+  enable_tf32: true
+  seed: 10086
+  num_train_epochs: 1
+  max_grad_norm: 1.0
+  method: random_masking
+  lower_p: 0.35
+  upper_p: 0.9
+  shrink: 8
+  post_num: 1
+  mask_times_per_sample: 2
+  max_gen_length: ${rollout.max_gen_length}
+  max_prompt_len: 512
+  eps: 0.2
+  beta: 0.01
+  use_kl_estimator_k3: true
+optimizer:
+  name: adamw
+  params:
+    learning_rate: 1.0e-06
+    scale_lr: false
+    beta1: 0.9
+    beta2: 0.999
+    weight_decay: 0.0
+    epsilon: 1.0e-08
+lr_scheduler:
+  scheduler: cosine
+  params:
+    learning_rate: ${optimizer.params.learning_rate}
+    warmup_steps: 0
+    min_lr_scale: 1.0
+evaluation:
+  if_eval: false
+  eval_dataset: video_reversal_rl_eval
+  data_type: video_reverse
+  num_response_per_task: 3
+  temperature: 0.1
+  steps: 128
+  max_gen_length: 128
+  batch_size: 1
+  remasking_strategy:
+  - low_confidence_static
+  - low_confidence_dynamic
+  target: confidence
+  dynamic_threshold: 0.9
+  block_size: 32
+  further_horizon: 128
+  use_cache: true
+config: configs/rl_lladav_video_reversal_rl.yaml

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step13.txt

@@ -0,0 +1,2 @@
+train step: 13  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.6367  avg length: 127.98
+train step: 13  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.6367  avg length: 127.98

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step15.txt

@@ -0,0 +1,2 @@
+train step: 15  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5391  avg length: 127.97
+train step: 15  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5391  avg length: 127.97

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step16.txt

@@ -0,0 +1,2 @@
+train step: 16  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5488  avg length: 127.99
+train step: 16  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5488  avg length: 127.99

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step18.txt

@@ -0,0 +1,2 @@
+train step: 18  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5332  avg length: 127.99
+train step: 18  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5332  avg length: 127.99

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step2.txt

@@ -0,0 +1,2 @@
+train step: 2  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5957  avg length: 127.99
+train step: 2  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5957  avg length: 127.99

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step21.txt

@@ -0,0 +1,2 @@
+train step: 21  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5938  avg length: 127.98
+train step: 21  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5938  avg length: 127.98

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step23.txt

@@ -0,0 +1,2 @@
+train step: 23  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5469  avg length: 127.97
+train step: 23  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5469  avg length: 127.97

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step24.txt

@@ -0,0 +1,2 @@
+train step: 24  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5352  avg length: 127.98
+train step: 24  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5352  avg length: 127.98

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step26.txt

@@ -0,0 +1,2 @@
+train step: 26  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5977  avg length: 127.99
+train step: 26  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5977  avg length: 127.99

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step29.txt

@@ -0,0 +1,2 @@
+train step: 29  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.4902  avg length: 127.98
+train step: 29  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.4902  avg length: 127.98

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step3.txt

@@ -0,0 +1,2 @@
+train step: 3  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5684  avg length: 128.00
+train step: 3  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.5684  avg length: 128.00

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step5.txt

@@ -0,0 +1,2 @@
+train step: 5  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.4199  avg length: 127.96
+train step: 5  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.4199  avg length: 127.96

rl_lladav_video_reversal_rl/results/results-rl-...rl_lladav_video_reversal_rl.ckpt.optimized-video_reversal_rl_train-step9.txt

@@ -0,0 +1,2 @@
+train step: 9  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.4492  avg length: 127.99
+train step: 9  remasking_strategy: low_confidence_dynamic  block_size: 32  acc: 0.4492  avg length: 127.99