aliangdw commited on
Commit
cf7160c
·
verified ·
1 Parent(s): d022de6

Update config.yaml

Browse files
Files changed (1) hide show
  1. config.yaml +0 -12
config.yaml CHANGED
@@ -18,8 +18,6 @@ custom_eval:
18
  reward_alignment:
19
  - libero_pi0_no_fail
20
  reward_alignment_max_trajectories: 10
21
- similarity_score:
22
- - aliangdw_metaworld_metaworld_eval
23
  subsample_n_frames: null
24
  use_frame_steps: true
25
  data:
@@ -46,8 +44,6 @@ data:
46
  max_trajectories: -1
47
  min_frames_per_trajectory: 1
48
  min_success: 0.5
49
- n_wrong_tasks: 5
50
- num_bins: 10
51
  partial_success_threshold: 0.2
52
  predict_last_frame_partial_progress: false
53
  preference_strategy_ratio:
@@ -72,14 +68,9 @@ data:
72
  seed: 42
73
  shuffle: true
74
  shuffle_progress_frames: false
75
- similarity_strategy_ratio:
76
- - 1.0
77
- - 1.0
78
- - 1.0
79
  train_datasets:
80
  - libero_pi0_no_fail
81
  traj_same_source_prob: 0.5
82
- use_data_source_balance: false
83
  use_multi_image: true
84
  use_per_frame_progress_token: true
85
  debug: false
@@ -131,7 +122,6 @@ model:
131
  train_language_model: true
132
  train_preference_head: true
133
  train_progress_head: true
134
- train_similarity_head: false
135
  train_success_head: false
136
  train_vision_encoder: false
137
  trust_remote_code: true
@@ -183,8 +173,6 @@ training:
183
  per_device_eval_batch_size: 64
184
  per_device_train_batch_size: 64
185
  predict_pref_progress: true
186
- predict_pref_sim: false
187
- predict_sim_progress: false
188
  prediction_loss_only: true
189
  remove_unused_columns: false
190
  resume_from_checkpoint: null
 
18
  reward_alignment:
19
  - libero_pi0_no_fail
20
  reward_alignment_max_trajectories: 10
 
 
21
  subsample_n_frames: null
22
  use_frame_steps: true
23
  data:
 
44
  max_trajectories: -1
45
  min_frames_per_trajectory: 1
46
  min_success: 0.5
 
 
47
  partial_success_threshold: 0.2
48
  predict_last_frame_partial_progress: false
49
  preference_strategy_ratio:
 
68
  seed: 42
69
  shuffle: true
70
  shuffle_progress_frames: false
 
 
 
 
71
  train_datasets:
72
  - libero_pi0_no_fail
73
  traj_same_source_prob: 0.5
 
74
  use_multi_image: true
75
  use_per_frame_progress_token: true
76
  debug: false
 
122
  train_language_model: true
123
  train_preference_head: true
124
  train_progress_head: true
 
125
  train_success_head: false
126
  train_vision_encoder: false
127
  trust_remote_code: true
 
173
  per_device_eval_batch_size: 64
174
  per_device_train_batch_size: 64
175
  predict_pref_progress: true
 
 
176
  prediction_loss_only: true
177
  remove_unused_columns: false
178
  resume_from_checkpoint: null