Add BiCAT 14-joint BTT tokenizer checkpoints at 100k and 150k
Browse files- BTT_tokenizer_step_100000/README.md +10 -0
- BTT_tokenizer_step_100000/config.json +56 -0
- BTT_tokenizer_step_100000/model.safetensors +3 -0
- BTT_tokenizer_step_100000/norm_stats.pt +3 -0
- BTT_tokenizer_step_100000/policy_io.json +15 -0
- BTT_tokenizer_step_100000/train_config.yaml +126 -0
- BTT_tokenizer_step_150000/README.md +10 -0
- BTT_tokenizer_step_150000/config.json +56 -0
- BTT_tokenizer_step_150000/model.safetensors +3 -0
- BTT_tokenizer_step_150000/norm_stats.pt +3 -0
- BTT_tokenizer_step_150000/policy_io.json +15 -0
- BTT_tokenizer_step_150000/train_config.yaml +126 -0
BTT_tokenizer_step_100000/README.md
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- model_hub_mixin
|
| 4 |
+
- pytorch_model_hub_mixin
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
|
| 8 |
+
- Code: [More Information Needed]
|
| 9 |
+
- Paper: [More Information Needed]
|
| 10 |
+
- Docs: [More Information Needed]
|
BTT_tokenizer_step_100000/config.json
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"action_dim": 14,
|
| 3 |
+
"action_horizon": 32,
|
| 4 |
+
"arm_dim": 7,
|
| 5 |
+
"clamp_normalized_actions": false,
|
| 6 |
+
"conv_kernel_sizes": [
|
| 7 |
+
5,
|
| 8 |
+
3
|
| 9 |
+
],
|
| 10 |
+
"conv_strides": [
|
| 11 |
+
2,
|
| 12 |
+
2
|
| 13 |
+
],
|
| 14 |
+
"decoder_nhead": 8,
|
| 15 |
+
"decoder_nlayer": 6,
|
| 16 |
+
"dropout": 0.1,
|
| 17 |
+
"fsq_levels": [
|
| 18 |
+
8,
|
| 19 |
+
8,
|
| 20 |
+
6,
|
| 21 |
+
5
|
| 22 |
+
],
|
| 23 |
+
"latent_dim": 4,
|
| 24 |
+
"latent_horizon_per_arm": 8,
|
| 25 |
+
"model_dim": 256,
|
| 26 |
+
"name": "btt",
|
| 27 |
+
"num_heads": 8,
|
| 28 |
+
"right_arm_sign": [
|
| 29 |
+
-1.0,
|
| 30 |
+
1.0,
|
| 31 |
+
1.0,
|
| 32 |
+
1.0,
|
| 33 |
+
-1.0,
|
| 34 |
+
-1.0,
|
| 35 |
+
1.0
|
| 36 |
+
],
|
| 37 |
+
"shared_action_max": [
|
| 38 |
+
0.84744793176651,
|
| 39 |
+
2.3744945526123047,
|
| 40 |
+
2.2520408630371094,
|
| 41 |
+
1.58407723903656,
|
| 42 |
+
1.3433661460876465,
|
| 43 |
+
2.050621747970581,
|
| 44 |
+
1.0
|
| 45 |
+
],
|
| 46 |
+
"shared_action_min": [
|
| 47 |
+
-0.7455939650535583,
|
| 48 |
+
-0.0036240178160369396,
|
| 49 |
+
0.001716639962978661,
|
| 50 |
+
-1.6111619472503662,
|
| 51 |
+
-0.7570382356643677,
|
| 52 |
+
-1.2132829427719116,
|
| 53 |
+
0.000721086747944355
|
| 54 |
+
],
|
| 55 |
+
"use_causal_decoder": true
|
| 56 |
+
}
|
BTT_tokenizer_step_100000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a106fc2fa9f5c2280d09438a0be20a252f528e96bd997899c78faceeb58d8317
|
| 3 |
+
size 29612324
|
BTT_tokenizer_step_100000/norm_stats.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:201d615d2c2dee4b0fc1713fb834ab7bdd45624bc05b7e36dd24564f7ab598bf
|
| 3 |
+
size 3957
|
BTT_tokenizer_step_100000/policy_io.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"obs_features": [],
|
| 3 |
+
"action_feature": {
|
| 4 |
+
"key": "action_joint_angles",
|
| 5 |
+
"type": "action",
|
| 6 |
+
"shape": [
|
| 7 |
+
14
|
| 8 |
+
],
|
| 9 |
+
"norm_mode": "min_max"
|
| 10 |
+
},
|
| 11 |
+
"n_obs_steps": 0,
|
| 12 |
+
"n_action_steps": 32,
|
| 13 |
+
"fps": 30,
|
| 14 |
+
"language_key": null
|
| 15 |
+
}
|
BTT_tokenizer_step_100000/train_config.yaml
ADDED
|
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
codec:
|
| 2 |
+
name: btt
|
| 3 |
+
kind: tokens
|
| 4 |
+
pretrained_path: null
|
| 5 |
+
action_dim: 14
|
| 6 |
+
arm_dim: 7
|
| 7 |
+
action_horizon: 32
|
| 8 |
+
latent_horizon_per_arm: 8
|
| 9 |
+
max_fit_chunks: 50000
|
| 10 |
+
model_dim: 256
|
| 11 |
+
num_heads: 8
|
| 12 |
+
decoder_nhead: 8
|
| 13 |
+
decoder_nlayer: 6
|
| 14 |
+
dropout: 0.1
|
| 15 |
+
use_causal_decoder: true
|
| 16 |
+
conv_kernel_sizes:
|
| 17 |
+
- 5
|
| 18 |
+
- 3
|
| 19 |
+
conv_strides:
|
| 20 |
+
- 2
|
| 21 |
+
- 2
|
| 22 |
+
latent_dim: 4
|
| 23 |
+
fsq_levels:
|
| 24 |
+
- 8
|
| 25 |
+
- 8
|
| 26 |
+
- 6
|
| 27 |
+
- 5
|
| 28 |
+
right_arm_sign:
|
| 29 |
+
- -1
|
| 30 |
+
- 1
|
| 31 |
+
- 1
|
| 32 |
+
- 1
|
| 33 |
+
- -1
|
| 34 |
+
- -1
|
| 35 |
+
- 1
|
| 36 |
+
shared_action_min:
|
| 37 |
+
- -0.7455939650535583
|
| 38 |
+
- -0.0036240178160369396
|
| 39 |
+
- 0.001716639962978661
|
| 40 |
+
- -1.6111619472503662
|
| 41 |
+
- -0.7570382356643677
|
| 42 |
+
- -1.2132829427719116
|
| 43 |
+
- 0.000721086747944355
|
| 44 |
+
shared_action_max:
|
| 45 |
+
- 0.84744793176651
|
| 46 |
+
- 2.3744945526123047
|
| 47 |
+
- 2.2520408630371094
|
| 48 |
+
- 1.58407723903656
|
| 49 |
+
- 1.3433661460876465
|
| 50 |
+
- 2.050621747970581
|
| 51 |
+
- 1.0
|
| 52 |
+
clamp_normalized_actions: false
|
| 53 |
+
dataset:
|
| 54 |
+
name: bicat_joint_all
|
| 55 |
+
repo_id: local/bicat_yam_joint_all
|
| 56 |
+
root: /data/home/shailesh/Evan/bimanual_tokenizer/data/BiCAT-BimanualYAM-JointOnly
|
| 57 |
+
hf_datasets_cache: ${oc.env:HF_DATASETS_CACHE,null}
|
| 58 |
+
episodes: null
|
| 59 |
+
action_key: action_joint_angles
|
| 60 |
+
exclude_episodes_file: null
|
| 61 |
+
action_stats_file: null
|
| 62 |
+
observation_keys: []
|
| 63 |
+
video_backend: pyav
|
| 64 |
+
bypass_force_cache_sync: true
|
| 65 |
+
trainer:
|
| 66 |
+
max_steps: 200000
|
| 67 |
+
batch_size: 128
|
| 68 |
+
global_batch_size: 512
|
| 69 |
+
num_workers: 4
|
| 70 |
+
prefetch_factor: 4
|
| 71 |
+
use_episode_aware_sampler: true
|
| 72 |
+
drop_n_first_frames: 0
|
| 73 |
+
drop_n_last_frames: 0
|
| 74 |
+
weight_decay: 0.0
|
| 75 |
+
betas:
|
| 76 |
+
- 0.9
|
| 77 |
+
- 0.95
|
| 78 |
+
eps: 1.0e-08
|
| 79 |
+
lr_scheduler: constant
|
| 80 |
+
warmup_steps: 200
|
| 81 |
+
min_lr_ratio: 0.1
|
| 82 |
+
use_ema: true
|
| 83 |
+
ema_decay: 0.995
|
| 84 |
+
mixed_precision: bf16
|
| 85 |
+
gradient_accumulation_steps: 1
|
| 86 |
+
find_unused_parameters: false
|
| 87 |
+
max_grad_norm: 1.0
|
| 88 |
+
distributed_strategy: ddp
|
| 89 |
+
fsdp_min_num_params: 100000000
|
| 90 |
+
fsdp_reshard_after_forward: true
|
| 91 |
+
fsdp_cpu_offload: false
|
| 92 |
+
log_every: 200
|
| 93 |
+
val_every: 0
|
| 94 |
+
checkpoint_every: 1000
|
| 95 |
+
checkpoint_dir: ${output_dir}/checkpoints
|
| 96 |
+
local_log_path: ${output_dir}/train_metrics.jsonl
|
| 97 |
+
val_ratio: 0.0
|
| 98 |
+
split_seed: ${seed}
|
| 99 |
+
val_batch_size: null
|
| 100 |
+
max_val_batches: 100
|
| 101 |
+
checkpoint_candidates:
|
| 102 |
+
best:
|
| 103 |
+
metric_key: val_reconst_loss
|
| 104 |
+
mode: min
|
| 105 |
+
k: 0
|
| 106 |
+
recent:
|
| 107 |
+
k: 5
|
| 108 |
+
use_wandb: true
|
| 109 |
+
wandb_project: BiCAT_realworld
|
| 110 |
+
wandb_entity: evan_lxw-harbin-institute-of-technology
|
| 111 |
+
wandb_required: true
|
| 112 |
+
wandb_init_timeout: null
|
| 113 |
+
wandb_upload_eval_videos: true
|
| 114 |
+
wandb_max_eval_videos: 16
|
| 115 |
+
_trainer_type: codecs
|
| 116 |
+
lr: 5.0e-05
|
| 117 |
+
reconst_every: 1000
|
| 118 |
+
max_reconst_batches: 100
|
| 119 |
+
skip_codec_fit: false
|
| 120 |
+
name: train_codecs
|
| 121 |
+
seed: 42
|
| 122 |
+
device: cuda
|
| 123 |
+
run_name: BiCAT_all_BTT_14joint_200k_67292
|
| 124 |
+
output_dir: /data/home/shailesh/Evan/bimanual_tokenizer/outputs/BiCAT_all_BTT_200k/67292/train
|
| 125 |
+
run_mode: overwrite
|
| 126 |
+
resume_checkpoint: null
|
BTT_tokenizer_step_150000/README.md
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- model_hub_mixin
|
| 4 |
+
- pytorch_model_hub_mixin
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
|
| 8 |
+
- Code: [More Information Needed]
|
| 9 |
+
- Paper: [More Information Needed]
|
| 10 |
+
- Docs: [More Information Needed]
|
BTT_tokenizer_step_150000/config.json
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"action_dim": 14,
|
| 3 |
+
"action_horizon": 32,
|
| 4 |
+
"arm_dim": 7,
|
| 5 |
+
"clamp_normalized_actions": false,
|
| 6 |
+
"conv_kernel_sizes": [
|
| 7 |
+
5,
|
| 8 |
+
3
|
| 9 |
+
],
|
| 10 |
+
"conv_strides": [
|
| 11 |
+
2,
|
| 12 |
+
2
|
| 13 |
+
],
|
| 14 |
+
"decoder_nhead": 8,
|
| 15 |
+
"decoder_nlayer": 6,
|
| 16 |
+
"dropout": 0.1,
|
| 17 |
+
"fsq_levels": [
|
| 18 |
+
8,
|
| 19 |
+
8,
|
| 20 |
+
6,
|
| 21 |
+
5
|
| 22 |
+
],
|
| 23 |
+
"latent_dim": 4,
|
| 24 |
+
"latent_horizon_per_arm": 8,
|
| 25 |
+
"model_dim": 256,
|
| 26 |
+
"name": "btt",
|
| 27 |
+
"num_heads": 8,
|
| 28 |
+
"right_arm_sign": [
|
| 29 |
+
-1.0,
|
| 30 |
+
1.0,
|
| 31 |
+
1.0,
|
| 32 |
+
1.0,
|
| 33 |
+
-1.0,
|
| 34 |
+
-1.0,
|
| 35 |
+
1.0
|
| 36 |
+
],
|
| 37 |
+
"shared_action_max": [
|
| 38 |
+
0.84744793176651,
|
| 39 |
+
2.3744945526123047,
|
| 40 |
+
2.2520408630371094,
|
| 41 |
+
1.58407723903656,
|
| 42 |
+
1.3433661460876465,
|
| 43 |
+
2.050621747970581,
|
| 44 |
+
1.0
|
| 45 |
+
],
|
| 46 |
+
"shared_action_min": [
|
| 47 |
+
-0.7455939650535583,
|
| 48 |
+
-0.0036240178160369396,
|
| 49 |
+
0.001716639962978661,
|
| 50 |
+
-1.6111619472503662,
|
| 51 |
+
-0.7570382356643677,
|
| 52 |
+
-1.2132829427719116,
|
| 53 |
+
0.000721086747944355
|
| 54 |
+
],
|
| 55 |
+
"use_causal_decoder": true
|
| 56 |
+
}
|
BTT_tokenizer_step_150000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6d41505da29324073030343f8ed88555240e972f2beee2a0cdad3cff1ec82bdb
|
| 3 |
+
size 29612324
|
BTT_tokenizer_step_150000/norm_stats.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:201d615d2c2dee4b0fc1713fb834ab7bdd45624bc05b7e36dd24564f7ab598bf
|
| 3 |
+
size 3957
|
BTT_tokenizer_step_150000/policy_io.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"obs_features": [],
|
| 3 |
+
"action_feature": {
|
| 4 |
+
"key": "action_joint_angles",
|
| 5 |
+
"type": "action",
|
| 6 |
+
"shape": [
|
| 7 |
+
14
|
| 8 |
+
],
|
| 9 |
+
"norm_mode": "min_max"
|
| 10 |
+
},
|
| 11 |
+
"n_obs_steps": 0,
|
| 12 |
+
"n_action_steps": 32,
|
| 13 |
+
"fps": 30,
|
| 14 |
+
"language_key": null
|
| 15 |
+
}
|
BTT_tokenizer_step_150000/train_config.yaml
ADDED
|
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
codec:
|
| 2 |
+
name: btt
|
| 3 |
+
kind: tokens
|
| 4 |
+
pretrained_path: null
|
| 5 |
+
action_dim: 14
|
| 6 |
+
arm_dim: 7
|
| 7 |
+
action_horizon: 32
|
| 8 |
+
latent_horizon_per_arm: 8
|
| 9 |
+
max_fit_chunks: 50000
|
| 10 |
+
model_dim: 256
|
| 11 |
+
num_heads: 8
|
| 12 |
+
decoder_nhead: 8
|
| 13 |
+
decoder_nlayer: 6
|
| 14 |
+
dropout: 0.1
|
| 15 |
+
use_causal_decoder: true
|
| 16 |
+
conv_kernel_sizes:
|
| 17 |
+
- 5
|
| 18 |
+
- 3
|
| 19 |
+
conv_strides:
|
| 20 |
+
- 2
|
| 21 |
+
- 2
|
| 22 |
+
latent_dim: 4
|
| 23 |
+
fsq_levels:
|
| 24 |
+
- 8
|
| 25 |
+
- 8
|
| 26 |
+
- 6
|
| 27 |
+
- 5
|
| 28 |
+
right_arm_sign:
|
| 29 |
+
- -1
|
| 30 |
+
- 1
|
| 31 |
+
- 1
|
| 32 |
+
- 1
|
| 33 |
+
- -1
|
| 34 |
+
- -1
|
| 35 |
+
- 1
|
| 36 |
+
shared_action_min:
|
| 37 |
+
- -0.7455939650535583
|
| 38 |
+
- -0.0036240178160369396
|
| 39 |
+
- 0.001716639962978661
|
| 40 |
+
- -1.6111619472503662
|
| 41 |
+
- -0.7570382356643677
|
| 42 |
+
- -1.2132829427719116
|
| 43 |
+
- 0.000721086747944355
|
| 44 |
+
shared_action_max:
|
| 45 |
+
- 0.84744793176651
|
| 46 |
+
- 2.3744945526123047
|
| 47 |
+
- 2.2520408630371094
|
| 48 |
+
- 1.58407723903656
|
| 49 |
+
- 1.3433661460876465
|
| 50 |
+
- 2.050621747970581
|
| 51 |
+
- 1.0
|
| 52 |
+
clamp_normalized_actions: false
|
| 53 |
+
dataset:
|
| 54 |
+
name: bicat_joint_all
|
| 55 |
+
repo_id: local/bicat_yam_joint_all
|
| 56 |
+
root: /data/home/shailesh/Evan/bimanual_tokenizer/data/BiCAT-BimanualYAM-JointOnly
|
| 57 |
+
hf_datasets_cache: ${oc.env:HF_DATASETS_CACHE,null}
|
| 58 |
+
episodes: null
|
| 59 |
+
action_key: action_joint_angles
|
| 60 |
+
exclude_episodes_file: null
|
| 61 |
+
action_stats_file: null
|
| 62 |
+
observation_keys: []
|
| 63 |
+
video_backend: pyav
|
| 64 |
+
bypass_force_cache_sync: true
|
| 65 |
+
trainer:
|
| 66 |
+
max_steps: 200000
|
| 67 |
+
batch_size: 128
|
| 68 |
+
global_batch_size: 512
|
| 69 |
+
num_workers: 4
|
| 70 |
+
prefetch_factor: 4
|
| 71 |
+
use_episode_aware_sampler: true
|
| 72 |
+
drop_n_first_frames: 0
|
| 73 |
+
drop_n_last_frames: 0
|
| 74 |
+
weight_decay: 0.0
|
| 75 |
+
betas:
|
| 76 |
+
- 0.9
|
| 77 |
+
- 0.95
|
| 78 |
+
eps: 1.0e-08
|
| 79 |
+
lr_scheduler: constant
|
| 80 |
+
warmup_steps: 200
|
| 81 |
+
min_lr_ratio: 0.1
|
| 82 |
+
use_ema: true
|
| 83 |
+
ema_decay: 0.995
|
| 84 |
+
mixed_precision: bf16
|
| 85 |
+
gradient_accumulation_steps: 1
|
| 86 |
+
find_unused_parameters: false
|
| 87 |
+
max_grad_norm: 1.0
|
| 88 |
+
distributed_strategy: ddp
|
| 89 |
+
fsdp_min_num_params: 100000000
|
| 90 |
+
fsdp_reshard_after_forward: true
|
| 91 |
+
fsdp_cpu_offload: false
|
| 92 |
+
log_every: 200
|
| 93 |
+
val_every: 0
|
| 94 |
+
checkpoint_every: 1000
|
| 95 |
+
checkpoint_dir: ${output_dir}/checkpoints
|
| 96 |
+
local_log_path: ${output_dir}/train_metrics.jsonl
|
| 97 |
+
val_ratio: 0.0
|
| 98 |
+
split_seed: ${seed}
|
| 99 |
+
val_batch_size: null
|
| 100 |
+
max_val_batches: 100
|
| 101 |
+
checkpoint_candidates:
|
| 102 |
+
best:
|
| 103 |
+
metric_key: val_reconst_loss
|
| 104 |
+
mode: min
|
| 105 |
+
k: 0
|
| 106 |
+
recent:
|
| 107 |
+
k: 5
|
| 108 |
+
use_wandb: true
|
| 109 |
+
wandb_project: BiCAT_realworld
|
| 110 |
+
wandb_entity: evan_lxw-harbin-institute-of-technology
|
| 111 |
+
wandb_required: true
|
| 112 |
+
wandb_init_timeout: null
|
| 113 |
+
wandb_upload_eval_videos: true
|
| 114 |
+
wandb_max_eval_videos: 16
|
| 115 |
+
_trainer_type: codecs
|
| 116 |
+
lr: 5.0e-05
|
| 117 |
+
reconst_every: 1000
|
| 118 |
+
max_reconst_batches: 100
|
| 119 |
+
skip_codec_fit: false
|
| 120 |
+
name: train_codecs
|
| 121 |
+
seed: 42
|
| 122 |
+
device: cuda
|
| 123 |
+
run_name: BiCAT_all_BTT_14joint_200k_67292
|
| 124 |
+
output_dir: /data/home/shailesh/Evan/bimanual_tokenizer/outputs/BiCAT_all_BTT_200k/67292/train
|
| 125 |
+
run_mode: overwrite
|
| 126 |
+
resume_checkpoint: null
|