Rename checkpoint references to Bagel-Robotics-1 and document private evaluation access
Browse files
README.md
CHANGED
|
@@ -7,9 +7,9 @@ tags:
|
|
| 7 |
- robodojo
|
| 8 |
---
|
| 9 |
|
| 10 |
-
#
|
| 11 |
|
| 12 |
-
|
| 13 |
|
| 14 |
The video expert has **12,468,409,664 parameters**, 23.95% fewer than its recorded parent. Learned masks were converted into smaller FFN and cross-attention layers. Self-attention heads and the action expert are retained. No teacher or learned-mask module is required for inference. Parameter reduction is not a measured inference-speedup claim.
|
| 15 |
|
|
@@ -24,8 +24,10 @@ The release adds documentation and inference packaging without changing any chec
|
|
| 24 |
|
| 25 |
## Download the exact checkpoint
|
| 26 |
|
|
|
|
|
|
|
| 27 |
```bash
|
| 28 |
-
hf download bageldotcom/
|
| 29 |
--revision 2651631c5d08819088475e8778676f5cf6d9afcf \
|
| 30 |
--include video.pt action.pt compact_architecture.json dataset_stats.json COMPLETE.json validation.json \
|
| 31 |
--local-dir /absolute/path/to/checkpoints/cp1000
|
|
@@ -35,17 +37,17 @@ There are approximately 29.0 GB of checkpoint tensors. Keep all six files togeth
|
|
| 35 |
|
| 36 |
## Inference adapter and reevaluation
|
| 37 |
|
| 38 |
-
The standalone XPolicyLab adapter is provided under [`inference/`](inference/), at release tag `
|
| 39 |
|
| 40 |
```bash
|
| 41 |
-
hf download bageldotcom/
|
| 42 |
-
--revision
|
| 43 |
--include 'inference/**' \
|
| 44 |
-
--local-dir /absolute/path/to/
|
| 45 |
|
| 46 |
git clone https://github.com/XPolicyLab/XPolicyLab.git /absolute/path/to/workspace/XPolicyLab
|
| 47 |
git -C /absolute/path/to/workspace/XPolicyLab checkout 6baccd57ff958570d0db1588370289ee36c8bb70
|
| 48 |
-
python /absolute/path/to/
|
| 49 |
```
|
| 50 |
|
| 51 |
Follow the included [policy README](inference/policy/bagel_robotics_1/README.md) to install the Python 3.10 environment, download/verify assets and run both debug checks. Follow [REEVALUATION.md](inference/policy/bagel_robotics_1/REEVALUATION.md) for task, suite and three-seed commands. Install RoboDojo and simulator assets separately using its [official guide](https://robodojo-benchmark.com/doc/usage/install-and-download/).
|
|
|
|
| 7 |
- robodojo
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# Bagel-Robotics-1 — checkpoint 1000
|
| 11 |
|
| 12 |
+
Inference release of the **exact checkpoint 1000** used in the Bagel-Robotics-1 evaluation campaign. **The repository is currently private.** Evaluator access is being arranged using a dedicated fine-grained token with read access only to this repository, delivered through an agreed private channel. Credential provisioning and evaluator assignment remain pending. Public release is prepared for the later verified-leaderboard milestone and requires a bageldotcom administrator. This is distinct from Bagel-Robotics-1-Preview.
|
| 13 |
|
| 14 |
The video expert has **12,468,409,664 parameters**, 23.95% fewer than its recorded parent. Learned masks were converted into smaller FFN and cross-attention layers. Self-attention heads and the action expert are retained. No teacher or learned-mask module is required for inference. Parameter reduction is not a measured inference-speedup claim.
|
| 15 |
|
|
|
|
| 24 |
|
| 25 |
## Download the exact checkpoint
|
| 26 |
|
| 27 |
+
While the repository is private, supply your authorized evaluator credential securely through `HF_TOKEN` or your Hugging Face login. Do not put credentials in commands, public documents, or application text. The commands below require that access.
|
| 28 |
+
|
| 29 |
```bash
|
| 30 |
+
hf download bageldotcom/Bagel-Robotics-1 \
|
| 31 |
--revision 2651631c5d08819088475e8778676f5cf6d9afcf \
|
| 32 |
--include video.pt action.pt compact_architecture.json dataset_stats.json COMPLETE.json validation.json \
|
| 33 |
--local-dir /absolute/path/to/checkpoints/cp1000
|
|
|
|
| 37 |
|
| 38 |
## Inference adapter and reevaluation
|
| 39 |
|
| 40 |
+
The standalone XPolicyLab adapter is provided under [`inference/`](inference/), at release tag `bagel-robotics-1-cp1000-inference-v1`. No private research repository is required.
|
| 41 |
|
| 42 |
```bash
|
| 43 |
+
hf download bageldotcom/Bagel-Robotics-1 \
|
| 44 |
+
--revision bagel-robotics-1-cp1000-inference-v1 \
|
| 45 |
--include 'inference/**' \
|
| 46 |
+
--local-dir /absolute/path/to/Bagel-Robotics-1-release
|
| 47 |
|
| 48 |
git clone https://github.com/XPolicyLab/XPolicyLab.git /absolute/path/to/workspace/XPolicyLab
|
| 49 |
git -C /absolute/path/to/workspace/XPolicyLab checkout 6baccd57ff958570d0db1588370289ee36c8bb70
|
| 50 |
+
python /absolute/path/to/Bagel-Robotics-1-release/inference/install_adapter.py /absolute/path/to/workspace/XPolicyLab
|
| 51 |
```
|
| 52 |
|
| 53 |
Follow the included [policy README](inference/policy/bagel_robotics_1/README.md) to install the Python 3.10 environment, download/verify assets and run both debug checks. Follow [REEVALUATION.md](inference/policy/bagel_robotics_1/REEVALUATION.md) for task, suite and three-seed commands. Install RoboDojo and simulator assets separately using its [official guide](https://robodojo-benchmark.com/doc/usage/install-and-download/).
|
RELEASE.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"model_name": "Bagel-Robotics-1",
|
| 3 |
-
"release_name": "
|
| 4 |
-
"repository": "bageldotcom/
|
| 5 |
"checkpoint_revision": "2651631c5d08819088475e8778676f5cf6d9afcf",
|
| 6 |
"checkpoint_identity": "790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c",
|
| 7 |
"files": {
|
|
@@ -30,7 +30,7 @@
|
|
| 30 |
"size": 24937214042
|
| 31 |
}
|
| 32 |
},
|
| 33 |
-
"scope": "
|
| 34 |
"historical_provenance": "Original checkpoint revision and checksums preserved. Older commits retain original provenance and export records; the current release view contains inference materials.",
|
| 35 |
"official_leaderboard_result": false
|
| 36 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"model_name": "Bagel-Robotics-1",
|
| 3 |
+
"release_name": "Bagel-Robotics-1 checkpoint 1000",
|
| 4 |
+
"repository": "bageldotcom/Bagel-Robotics-1",
|
| 5 |
"checkpoint_revision": "2651631c5d08819088475e8778676f5cf6d9afcf",
|
| 6 |
"checkpoint_identity": "790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c",
|
| 7 |
"files": {
|
|
|
|
| 30 |
"size": 24937214042
|
| 31 |
}
|
| 32 |
},
|
| 33 |
+
"scope": "Private inference checkpoint and supporting adapter for evaluator access; public release prepared; training/pruning/export source planned November 20, 2026",
|
| 34 |
"historical_provenance": "Original checkpoint revision and checksums preserved. Older commits retain original provenance and export records; the current release view contains inference materials.",
|
| 35 |
"official_leaderboard_result": false
|
| 36 |
}
|
artifact-manifest.json
CHANGED
|
@@ -4,9 +4,9 @@
|
|
| 4 |
"checkpoint_identity": "790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c",
|
| 5 |
"update": 1000,
|
| 6 |
"video_parameters": 12468409664,
|
| 7 |
-
"repo_id": "bageldotcom/
|
| 8 |
"revision": "2651631c5d08819088475e8778676f5cf6d9afcf",
|
| 9 |
-
"private":
|
| 10 |
"files": {
|
| 11 |
"COMPLETE.json": {
|
| 12 |
"sha256": "eee062fbcb8c92a881117aa7f06b5cd039dfb9912d85571b59a20aa30837b0e1",
|
|
|
|
| 4 |
"checkpoint_identity": "790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c",
|
| 5 |
"update": 1000,
|
| 6 |
"video_parameters": 12468409664,
|
| 7 |
+
"repo_id": "bageldotcom/Bagel-Robotics-1",
|
| 8 |
"revision": "2651631c5d08819088475e8778676f5cf6d9afcf",
|
| 9 |
+
"private": true,
|
| 10 |
"files": {
|
| 11 |
"COMPLETE.json": {
|
| 12 |
"sha256": "eee062fbcb8c92a881117aa7f06b5cd039dfb9912d85571b59a20aa30837b0e1",
|
inference/README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
-
# Bagel-Robotics-1
|
| 2 |
|
| 3 |
-
This folder is the standalone XPolicyLab adapter for
|
| 4 |
|
| 5 |
Install this adapter into an existing XPolicyLab checkout (the installer refuses to overwrite an existing policy):
|
| 6 |
|
|
@@ -10,4 +10,6 @@ python install_adapter.py /absolute/path/to/XPolicyLab
|
|
| 10 |
|
| 11 |
Then follow [Installation](policy/bagel_robotics_1/README.md#installation) for the Python 3.10 environment, exact checkpoint and encoder download, and debug checks. [REEVALUATION.md](policy/bagel_robotics_1/REEVALUATION.md) gives native simulator task/suite/three-seed commands. RoboDojo and its assets are installed separately.
|
| 12 |
|
| 13 |
-
The adapter is prepared against XPolicyLab `6baccd57ff958570d0db1588370289ee36c8bb70`; source and dependency pins are preserved in the policy folder. This
|
|
|
|
|
|
|
|
|
| 1 |
+
# Bagel-Robotics-1 inference adapter
|
| 2 |
|
| 3 |
+
This folder is the standalone XPolicyLab adapter for Bagel-Robotics-1 checkpoint 1000. It contains inference code and evaluation instructions, with no training repository dependency. Model tensors download separately from `bageldotcom/Bagel-Robotics-1`, pinned to `2651631c5d08819088475e8778676f5cf6d9afcf` and verified against `policy/bagel_robotics_1/artifact-manifest.json`.
|
| 4 |
|
| 5 |
Install this adapter into an existing XPolicyLab checkout (the installer refuses to overwrite an existing policy):
|
| 6 |
|
|
|
|
| 10 |
|
| 11 |
Then follow [Installation](policy/bagel_robotics_1/README.md#installation) for the Python 3.10 environment, exact checkpoint and encoder download, and debug checks. [REEVALUATION.md](policy/bagel_robotics_1/REEVALUATION.md) gives native simulator task/suite/three-seed commands. RoboDojo and its assets are installed separately.
|
| 12 |
|
| 13 |
+
The adapter is prepared against XPolicyLab `6baccd57ff958570d0db1588370289ee36c8bb70`; source and dependency pins are preserved in the policy folder. This inference package is not an XPolicyLab PR or an official leaderboard result. Training/pruning/export source and the technical report are planned for November 20, 2026.
|
| 14 |
+
|
| 15 |
+
The repository is currently private. Evaluators need separately arranged repository-only read access; public release is a later leaderboard milestone.
|
inference/SHA256SUMS
CHANGED
|
@@ -1,14 +1,14 @@
|
|
| 1 |
-
|
| 2 |
56f26e5b6ba87080c82703e102240fdfbd955c731fb2a044609ffff099795b3f install_adapter.py
|
| 3 |
6c9437916a0dc5b6ed5fd56b361a4a1894be910e6965c9570e8683cc74a40813 policy/bagel_robotics_1/.gitignore
|
| 4 |
24cb153ee8bababbfc20fe7acd36768923fbcf2a07d0aa653f460282761ccf8f policy/bagel_robotics_1/EVALUATION_RESULTS.md
|
| 5 |
8379dddfbbb1154ad9cc57f008f1f8d4a8d26c3de99bd6b36aa4d8067888aa92 policy/bagel_robotics_1/LICENSE
|
| 6 |
-
|
| 7 |
-
|
| 8 |
2f051521079a9de8815faafcb20098f1fe7fa167425e8c8088e7a42c2d50bb8e policy/bagel_robotics_1/THIRD_PARTY_NOTICES.md
|
| 9 |
1d22bc8164fcca87f47b003406878112087118c4a5e23e6019ae1998e2b53418 policy/bagel_robotics_1/VALIDATION.md
|
| 10 |
e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 policy/bagel_robotics_1/__init__.py
|
| 11 |
-
|
| 12 |
76fec0d61c9344d8d2c5f61853a93269c8b19ae16d08de673f2ced5bdfe08b79 policy/bagel_robotics_1/assets.py
|
| 13 |
2a16b03892ede7aa0a93cfd65e01a239498aa992bc4f1affb9cdde54253dae08 policy/bagel_robotics_1/backend.py
|
| 14 |
68f337b21b1e517e1df27550e64fbc542b644528adc8f3f47fb8e44849185af6 policy/bagel_robotics_1/deploy.py
|
|
|
|
| 1 |
+
9b146bf4e4b90a6bc16f5e904ee7b6a195e2ab99b7a1b2b6169dcbc614714182 README.md
|
| 2 |
56f26e5b6ba87080c82703e102240fdfbd955c731fb2a044609ffff099795b3f install_adapter.py
|
| 3 |
6c9437916a0dc5b6ed5fd56b361a4a1894be910e6965c9570e8683cc74a40813 policy/bagel_robotics_1/.gitignore
|
| 4 |
24cb153ee8bababbfc20fe7acd36768923fbcf2a07d0aa653f460282761ccf8f policy/bagel_robotics_1/EVALUATION_RESULTS.md
|
| 5 |
8379dddfbbb1154ad9cc57f008f1f8d4a8d26c3de99bd6b36aa4d8067888aa92 policy/bagel_robotics_1/LICENSE
|
| 6 |
+
430474f9f20a388569d7d848a03c97b1fde55fcdaec8efe4c95db663f895789d policy/bagel_robotics_1/README.md
|
| 7 |
+
3efd04c891dc86f4742e5a70a6ac19beb71c0d1a4767549afdf7f2dc87d791fd policy/bagel_robotics_1/REEVALUATION.md
|
| 8 |
2f051521079a9de8815faafcb20098f1fe7fa167425e8c8088e7a42c2d50bb8e policy/bagel_robotics_1/THIRD_PARTY_NOTICES.md
|
| 9 |
1d22bc8164fcca87f47b003406878112087118c4a5e23e6019ae1998e2b53418 policy/bagel_robotics_1/VALIDATION.md
|
| 10 |
e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 policy/bagel_robotics_1/__init__.py
|
| 11 |
+
e7667cec0939b219dc1396dcf34c8fadcd95f509e6606df9757b6c909967b5f2 policy/bagel_robotics_1/artifact-manifest.json
|
| 12 |
76fec0d61c9344d8d2c5f61853a93269c8b19ae16d08de673f2ced5bdfe08b79 policy/bagel_robotics_1/assets.py
|
| 13 |
2a16b03892ede7aa0a93cfd65e01a239498aa992bc4f1affb9cdde54253dae08 policy/bagel_robotics_1/backend.py
|
| 14 |
68f337b21b1e517e1df27550e64fbc542b644528adc8f3f47fb8e44849185af6 policy/bagel_robotics_1/deploy.py
|
inference/policy/bagel_robotics_1/README.md
CHANGED
|
@@ -38,9 +38,15 @@ establish numerical equivalence. The recorded GPU has 96 GB VRAM; a lower
|
|
| 38 |
minimum has not been qualified. The isolated validation jobs reserved 128 GiB
|
| 39 |
host RAM and eight CPUs; these are tested allocations, not measured minima.
|
| 40 |
|
| 41 |
-
**2. Download and verify before evaluation.** The
|
| 42 |
-
[bageldotcom/
|
| 43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
|
| 45 |
```bash
|
| 46 |
PYTHON_BIN="$BAGEL_POLICY_ENV/bin/python" bash download_checkpoint.sh
|
|
@@ -153,8 +159,8 @@ The Installation quickstart authenticates and downloads before evaluation.
|
|
| 153 |
The script downloads the six exact bundle files into `checkpoints/cp1000` and
|
| 154 |
the pinned public VAE/CLIP/UMT5/tokenizer assets into
|
| 155 |
`checkpoints/Wan2.1-I2V-14B-480P`. Bundle repository:
|
| 156 |
-
[bageldotcom/
|
| 157 |
-
revision `2651631c5d08819088475e8778676f5cf6d9afcf` (**public;
|
| 158 |
Access must be arranged for evaluators. No token is included.
|
| 159 |
|
| 160 |
For existing artifacts, set `BAGEL_CP1000_BUNDLE` and `BAGEL_CP1000_WAN_ROOT`.
|
|
|
|
| 38 |
minimum has not been qualified. The isolated validation jobs reserved 128 GiB
|
| 39 |
host RAM and eight CPUs; these are tested allocations, not measured minima.
|
| 40 |
|
| 41 |
+
**2. Download and verify before evaluation.** The public inference release is
|
| 42 |
+
prepared at [bageldotcom/Bagel-Robotics-1](https://huggingface.co/bageldotcom/Bagel-Robotics-1).
|
| 43 |
+
The repository is currently private. Evaluators need a dedicated fine-grained
|
| 44 |
+
token granting read access only to this repository, supplied separately through
|
| 45 |
+
an agreed private channel. Credential creation, validation and handoff are pending.
|
| 46 |
+
Set `HF_TOKEN` securely before downloading (for example, `read -r -s HF_TOKEN`
|
| 47 |
+
followed by `export HF_TOKEN`; use `set +x` first and `unset HF_TOKEN` afterward).
|
| 48 |
+
Do not place the token in scripts, the PR, logs, or the application text.
|
| 49 |
+
Public release is a separate leaderboard milestone; once public, no token is needed.
|
| 50 |
|
| 51 |
```bash
|
| 52 |
PYTHON_BIN="$BAGEL_POLICY_ENV/bin/python" bash download_checkpoint.sh
|
|
|
|
| 159 |
The script downloads the six exact bundle files into `checkpoints/cp1000` and
|
| 160 |
the pinned public VAE/CLIP/UMT5/tokenizer assets into
|
| 161 |
`checkpoints/Wan2.1-I2V-14B-480P`. Bundle repository:
|
| 162 |
+
[bageldotcom/Bagel-Robotics-1](https://huggingface.co/bageldotcom/Bagel-Robotics-1),
|
| 163 |
+
revision `2651631c5d08819088475e8778676f5cf6d9afcf` (**public release prepared; visibility change pending**).
|
| 164 |
Access must be arranged for evaluators. No token is included.
|
| 165 |
|
| 166 |
For existing artifacts, set `BAGEL_CP1000_BUNDLE` and `BAGEL_CP1000_WAN_ROOT`.
|
inference/policy/bagel_robotics_1/REEVALUATION.md
CHANGED
|
@@ -6,11 +6,11 @@ debug checks. The standalone inference runtime is included; the private research
|
|
| 6 |
training repository is not needed. Install RoboDojo and its assets separately.
|
| 7 |
|
| 8 |
The only policy checkpoint is `cp1000`, from
|
| 9 |
-
[bageldotcom/
|
| 10 |
revision `2651631c5d08819088475e8778676f5cf6d9afcf`. The downloader and loader reject
|
| 11 |
files that do not match `artifact-manifest.json`. The VAE, image/text encoders and
|
| 12 |
tokenizer are downloaded automatically at their separately pinned public revision.
|
| 13 |
-
The checkpoint
|
| 14 |
|
| 15 |
## Set the local paths
|
| 16 |
|
|
|
|
| 6 |
training repository is not needed. Install RoboDojo and its assets separately.
|
| 7 |
|
| 8 |
The only policy checkpoint is `cp1000`, from
|
| 9 |
+
[bageldotcom/Bagel-Robotics-1](https://huggingface.co/bageldotcom/Bagel-Robotics-1),
|
| 10 |
revision `2651631c5d08819088475e8778676f5cf6d9afcf`. The downloader and loader reject
|
| 11 |
files that do not match `artifact-manifest.json`. The VAE, image/text encoders and
|
| 12 |
tokenizer are downloaded automatically at their separately pinned public revision.
|
| 13 |
+
The checkpoint is prepared for public access, pending an organization administrator visibility change. Once public, it and the required encoder assets download without authentication.
|
| 14 |
|
| 15 |
## Set the local paths
|
| 16 |
|
inference/policy/bagel_robotics_1/artifact-manifest.json
CHANGED
|
@@ -4,9 +4,9 @@
|
|
| 4 |
"checkpoint_identity": "790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c",
|
| 5 |
"update": 1000,
|
| 6 |
"video_parameters": 12468409664,
|
| 7 |
-
"repo_id": "bageldotcom/
|
| 8 |
"revision": "2651631c5d08819088475e8778676f5cf6d9afcf",
|
| 9 |
-
"private":
|
| 10 |
"files": {
|
| 11 |
"COMPLETE.json": {
|
| 12 |
"sha256": "eee062fbcb8c92a881117aa7f06b5cd039dfb9912d85571b59a20aa30837b0e1",
|
|
|
|
| 4 |
"checkpoint_identity": "790bffc9464fa7a026aa43b217aec2033d186895cad643166daa0bcc01914d5c",
|
| 5 |
"update": 1000,
|
| 6 |
"video_parameters": 12468409664,
|
| 7 |
+
"repo_id": "bageldotcom/Bagel-Robotics-1",
|
| 8 |
"revision": "2651631c5d08819088475e8778676f5cf6d9afcf",
|
| 9 |
+
"private": true,
|
| 10 |
"files": {
|
| 11 |
"COMPLETE.json": {
|
| 12 |
"sha256": "eee062fbcb8c92a881117aa7f06b5cd039dfb9912d85571b59a20aa30837b0e1",
|