harryzhang1018 commited on
Commit
8a7a59e
·
verified ·
1 Parent(s): 6039542

SO101: point study landing page to pinned external source and results summary

Browse files
Files changed (1) hide show
  1. so101_pick_place/README.md +12 -6
so101_pick_place/README.md CHANGED
@@ -13,11 +13,14 @@ wrist-roll limit of 98 deg (T5 / S4: from the motor range and two robot stalls)
13
  a fitted servo model (used only in the sim checks); and the choice of the student for the robot. Each fine-tune round
14
  (S3, T5 / S4, T6 / S5) answered failures of the student before it on the robot.
15
 
16
- The code, the configs and the commands are in the folders `so101_pick_place/` and `scripts/so101_pick_place/` of
17
- the main branch of the code repository [uwsbel/NeDM](https://github.com/uwsbel/NeDM/tree/main/so101_pick_place) (merged from the branch `so101-pick-place`). The
18
- study history, with every stage commit named below, is the branch `so101-pick-place-study` (tip
19
- `3599103704032ba67cb2736ff0aee501f2e537b9`). Sizes and SHA-256 of every file: `release_manifest.json` (it also holds
20
- the provenance of each folder).
 
 
 
21
 
22
  ## Results on the real arm
23
 
@@ -59,6 +62,7 @@ Per-episode table: `results/robot/results.md` (built from `robot/logs/` by `scri
59
 
60
  | Path | Contents | Size |
61
  |---|---|---|
 
62
  | `nrd/c1_s62/` | The NRD of every policy stage. `best.pt`: main-compatible copy (note 2). `original/best.pt`: the file the study used (SHA-256 `f45ee8be...`). Training records: `run_config.json`, `train_log.jsonl`, `validation.json`, `complete.json`, `contact_scale.json`, `core_validation.json` | 37 MB |
63
  | `nrd/c0_s62/` | Base NRD (the start weights of `c1_s62`); same files | 28 MB |
64
  | `nrd_data/production_plus/` | 37,500 scripted MuJoCo episodes (12 grasp families), the base NRD's training data | 4.24 GB |
@@ -145,8 +149,10 @@ repeat bit for bit: a re-run gives a similar, not an equal, checkpoint.
145
 
146
  ## Download
147
 
 
 
148
  ```bash
149
- scripts/so101_pick_place/reproduce.sh smoke --steps download # in the code repository; checks size and SHA-256 of every file
150
  ```
151
 
152
  Or with `huggingface_hub`: `snapshot_download("harryzhang1018/NeDM", repo_type="dataset", revision=<hf_revision>,
 
13
  a fitted servo model (used only in the sim checks); and the choice of the student for the robot. Each fine-tune round
14
  (S3, T5 / S4, T6 / S5) answered failures of the student before it on the robot.
15
 
16
+ The study implementation, configs, requirements and full reproduction instructions are now hosted in the
17
+ [pinned source folder](https://huggingface.co/datasets/harryzhang1018/NeDM/tree/6039542eea5eb96bd925f64dcdaf94d647d5d67c/so101_pick_place/code).
18
+ Start with its [reproduction guide](https://huggingface.co/datasets/harryzhang1018/NeDM/blob/6039542eea5eb96bd925f64dcdaf94d647d5d67c/so101_pick_place/code/REPRODUCE.md).
19
+ The NeDM repository keeps the [result summary and compact release manifest](https://github.com/uwsbel/NeDM/tree/main/so101_pick_place).
20
+ The original 503-file data manifest remains pinned at `dcb873bc5da3da8ad5abdec6137c77ef46038f43`; source and
21
+ payload revisions are recorded separately in the compact manifest. Checkpoints, datasets, logs and videos retain
22
+ their original bytes. The Git study history remains on `so101-pick-place-study`
23
+ (tip `3599103704032ba67cb2736ff0aee501f2e537b9`).
24
 
25
  ## Results on the real arm
26
 
 
62
 
63
  | Path | Contents | Size |
64
  |---|---|---|
65
+ | `code/` | Study source, configs, requirements, full original README, checksums and source overlay; start with `REPRODUCE.md` | 3 MB |
66
  | `nrd/c1_s62/` | The NRD of every policy stage. `best.pt`: main-compatible copy (note 2). `original/best.pt`: the file the study used (SHA-256 `f45ee8be...`). Training records: `run_config.json`, `train_log.jsonl`, `validation.json`, `complete.json`, `contact_scale.json`, `core_validation.json` | 37 MB |
67
  | `nrd/c0_s62/` | Base NRD (the start weights of `c1_s62`); same files | 28 MB |
68
  | `nrd_data/production_plus/` | 37,500 scripted MuJoCo episodes (12 grasp families), the base NRD's training data | 4.24 GB |
 
149
 
150
  ## Download
151
 
152
+ Restore the pinned source checkout as described above before running these commands.
153
+
154
  ```bash
155
+ scripts/so101_pick_place/reproduce.sh smoke --steps download # in the restored source checkout; checks size and SHA-256 of every file
156
  ```
157
 
158
  Or with `huggingface_hub`: `snapshot_download("harryzhang1018/NeDM", repo_type="dataset", revision=<hf_revision>,