rsj2003 commited on
Commit
0b5a17c
·
verified ·
1 Parent(s): 91a5167

Upload StreamingWAM LIBERO checkpoint

Browse files
Files changed (3) hide show
  1. README.md +51 -0
  2. checkpoint_info.json +20 -0
  3. streamingwam_libero.pt +3 -0
README.md ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ tags:
4
+ - robotics
5
+ - imitation-learning
6
+ - diffusion-policy
7
+ - libero
8
+ ---
9
+
10
+ # StreamingWAM LIBERO checkpoint
11
+
12
+ This repository contains the released StreamingWAM checkpoint trained on the
13
+ four standard LIBERO suites and evaluated on LIBERO-Plus. The checkpoint is
14
+ intended for use with the accompanying
15
+ [StreamingWAM code](https://github.com/renshaojie233/StreamingWAM).
16
+
17
+ ## File
18
+
19
+ | File | Size | SHA-256 |
20
+ | --- | ---: | --- |
21
+ | `streamingwam_libero.pt` | 12,042,077,420 bytes | `35499c8b2ac7bc879c988c9af4f9e9ff9052caccd22d582b7fadd90de185d496` |
22
+
23
+ The PyTorch checkpoint contains the model weights (`mot` and
24
+ `proprio_encoder`) and minimal loading metadata (`step` and `torch_dtype`). It
25
+ does not contain optimizer state, scheduler state, random-number-generator
26
+ state, training logs, or experiment-tracking data.
27
+
28
+ ## Usage
29
+
30
+ Download `streamingwam_libero.pt`, install the StreamingWAM codebase, and pass
31
+ the local checkpoint path to the evaluation command:
32
+
33
+ ```bash
34
+ python experiments/libero/eval_libero_task_list_multi_k.py \
35
+ task=streamingwam_libero_plus \
36
+ ckpt=/path/to/streamingwam_libero.pt \
37
+ EVALUATION.dataset_stats_path=assets/libero_dataset_stats.json \
38
+ +EVALUATION.task_list_file=assets/libero_plus_full_10030.txt \
39
+ +EVALUATION.sdp_k_values='[4]'
40
+ ```
41
+
42
+ See the code repository for the complete environment, data layout, and
43
+ evaluation options.
44
+
45
+ ## Scope and terms
46
+
47
+ This checkpoint is released for research on robot learning and policy
48
+ evaluation. The source code is MIT-licensed. External datasets, pretrained
49
+ components, benchmark assets, and model weights retain their respective
50
+ licenses and terms.
51
+
checkpoint_info.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "file": "streamingwam_libero.pt",
3
+ "format": "PyTorch checkpoint",
4
+ "size_bytes": 12042077420,
5
+ "sha256": "35499c8b2ac7bc879c988c9af4f9e9ff9052caccd22d582b7fadd90de185d496",
6
+ "training_benchmark": "LIBERO",
7
+ "evaluation_benchmark": "LIBERO-Plus",
8
+ "tensor_groups": {
9
+ "mot": 1649,
10
+ "proprio_encoder": 2
11
+ },
12
+ "torch_dtype": "torch.bfloat16",
13
+ "excluded": [
14
+ "optimizer state",
15
+ "scheduler state",
16
+ "random-number-generator state",
17
+ "training logs",
18
+ "experiment-tracking data"
19
+ ]
20
+ }
streamingwam_libero.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35499c8b2ac7bc879c988c9af4f9e9ff9052caccd22d582b7fadd90de185d496
3
+ size 12042077420