Use complete VLA checkpoint; remove redundant Qwen weights
Browse files- README.md +3 -3
- SHA256SUMS +2 -3
- base_vlm/Qwen3.5-4B-Action/model.safetensors +0 -3
- release.json +9 -2
README.md
CHANGED
|
@@ -26,7 +26,7 @@ MR0 continuous-action pretraining model with a Qwen3.5-4B vision-language backbo
|
|
| 26 |
|
| 27 |
## Download and load
|
| 28 |
|
| 29 |
-
Install the MR0 code
|
| 30 |
|
| 31 |
```python
|
| 32 |
import os
|
|
@@ -45,9 +45,9 @@ model = baseframework.from_pretrained(root / "checkpoints/model.pt")
|
|
| 45 |
model = model.eval().to("cuda")
|
| 46 |
```
|
| 47 |
|
| 48 |
-
|
| 49 |
|
| 50 |
-
The checkpoint bytes are unchanged. `release.json` records the model provenance and `SHA256SUMS` records file identities. For downstream training, adapt the dataset settings to
|
| 51 |
|
| 52 |
## Acknowledgements
|
| 53 |
|
|
|
|
| 26 |
|
| 27 |
## Download and load
|
| 28 |
|
| 29 |
+
Install the MR0 code at commit [`88c66793`](https://github.com/meituan/Meituan-Robotics-0/commit/88c667933dfc941a71c9b142eb84b716ac4f0707) or later, following the Getting Started guide. This loader reads the complete VLA checkpoint without a separate Qwen weight file. Then download this repository:
|
| 30 |
|
| 31 |
```python
|
| 32 |
import os
|
|
|
|
| 45 |
model = model.eval().to("cuda")
|
| 46 |
```
|
| 47 |
|
| 48 |
+
`checkpoints/model.pt` contains the complete Qwen3.5 backbone and MMDiT action head. `base_vlm/Qwen3.5-4B-Action/` contains only the matching tokenizer, processor, and architecture configuration; it is not a second model. The release also includes the VLA configuration and normalization statistics. Choose the normalization key in `dataset_statistics.json` that matches your robot embodiment.
|
| 49 |
|
| 50 |
+
The checkpoint bytes are unchanged. `release.json` records the model provenance and `SHA256SUMS` records file identities. For downstream training, adapt the dataset settings and set both `framework.qwenvl.checkpoint_path` and `trainer.pretrained_checkpoint` to the downloaded complete checkpoint. Set `framework.qwenvl.base_vlm` to its metadata directory.
|
| 51 |
|
| 52 |
## Acknowledgements
|
| 53 |
|
SHA256SUMS
CHANGED
|
@@ -1,13 +1,12 @@
|
|
| 1 |
-
|
| 2 |
0eebf5948dad3e1b23783277a80eb88293b1f3182c7f0753533b95d34b3e6fda base_vlm/Qwen3.5-4B-Action/added_custom_token_id_map.json
|
| 3 |
a4aee8afcf2e0711942cf848899be66016f8d14a889ff9ede07bca099c28f715 base_vlm/Qwen3.5-4B-Action/chat_template.jinja
|
| 4 |
65381958191ecc78ce7669a2146e882d020352f636bab38a7681b0754cce0dd1 base_vlm/Qwen3.5-4B-Action/config.json
|
| 5 |
ffe718fead008893ae749d52595ca89f1325ae2800f7eb5beb4bea4b8913883d base_vlm/Qwen3.5-4B-Action/generation_config.json
|
| 6 |
-
89eee75b9962a972723c76bfa3ca5d762a11383e9aad15dde1acff786f12dab0 base_vlm/Qwen3.5-4B-Action/model.safetensors
|
| 7 |
d89ef49ce9cd37fbf510158e13c1ef063d9286411c1ec9049932dbe0487143b1 base_vlm/Qwen3.5-4B-Action/processor_config.json
|
| 8 |
fa858c43fc03e4fefc9028af319b9bfe3aa3623e8833e5ed4a1890bfdd4daa27 base_vlm/Qwen3.5-4B-Action/tokenizer.json
|
| 9 |
55c5979db375311c4989112e1223de2f994c7b97ba04f9e2e15579b9ee60a87b base_vlm/Qwen3.5-4B-Action/tokenizer_config.json
|
| 10 |
2d61674add3196244984aae7a08c069b270d0b2b28c69aa5655e37e720930446 checkpoints/model.pt
|
| 11 |
b004a6bc1a49e0fd0b520f64074149401fe872fec276e68ebdd71d6f984a583e config.full.yaml
|
| 12 |
791f37803e18eb16cdf18c588bee76f4a3258a09d0875e2f4d19652d2d7cf4b1 dataset_statistics.json
|
| 13 |
-
|
|
|
|
| 1 |
+
5d8dbde6ea2d6bda06d53561124b8391d271b6f3d38c61e2b2f1db0ddcf0f0dc README.md
|
| 2 |
0eebf5948dad3e1b23783277a80eb88293b1f3182c7f0753533b95d34b3e6fda base_vlm/Qwen3.5-4B-Action/added_custom_token_id_map.json
|
| 3 |
a4aee8afcf2e0711942cf848899be66016f8d14a889ff9ede07bca099c28f715 base_vlm/Qwen3.5-4B-Action/chat_template.jinja
|
| 4 |
65381958191ecc78ce7669a2146e882d020352f636bab38a7681b0754cce0dd1 base_vlm/Qwen3.5-4B-Action/config.json
|
| 5 |
ffe718fead008893ae749d52595ca89f1325ae2800f7eb5beb4bea4b8913883d base_vlm/Qwen3.5-4B-Action/generation_config.json
|
|
|
|
| 6 |
d89ef49ce9cd37fbf510158e13c1ef063d9286411c1ec9049932dbe0487143b1 base_vlm/Qwen3.5-4B-Action/processor_config.json
|
| 7 |
fa858c43fc03e4fefc9028af319b9bfe3aa3623e8833e5ed4a1890bfdd4daa27 base_vlm/Qwen3.5-4B-Action/tokenizer.json
|
| 8 |
55c5979db375311c4989112e1223de2f994c7b97ba04f9e2e15579b9ee60a87b base_vlm/Qwen3.5-4B-Action/tokenizer_config.json
|
| 9 |
2d61674add3196244984aae7a08c069b270d0b2b28c69aa5655e37e720930446 checkpoints/model.pt
|
| 10 |
b004a6bc1a49e0fd0b520f64074149401fe872fec276e68ebdd71d6f984a583e config.full.yaml
|
| 11 |
791f37803e18eb16cdf18c588bee76f4a3258a09d0875e2f4d19652d2d7cf4b1 dataset_statistics.json
|
| 12 |
+
75f607fcb7416f0453d71ebcc3ebfd09901f5d6e8d197ea79a455a3e0da3201e release.json
|
base_vlm/Qwen3.5-4B-Action/model.safetensors
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:89eee75b9962a972723c76bfa3ca5d762a11383e9aad15dde1acff786f12dab0
|
| 3 |
-
size 9087862136
|
|
|
|
|
|
|
|
|
|
|
|
release.json
CHANGED
|
@@ -7,8 +7,15 @@
|
|
| 7 |
"action_dim": 20,
|
| 8 |
"action_horizon": 16,
|
| 9 |
"code_repository": "https://github.com/meituan/Meituan-Robotics-0",
|
| 10 |
-
"code_commit": "
|
| 11 |
"serialization": "PyTorch state_dict; unchanged from the selected source checkpoint",
|
| 12 |
"initialization": "Stage 1 FAST checkpoint at 430000 training steps",
|
| 13 |
-
"repo_id": "MeituanRobotics/Meituan-Robotics-0"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
}
|
|
|
|
| 7 |
"action_dim": 20,
|
| 8 |
"action_horizon": 16,
|
| 9 |
"code_repository": "https://github.com/meituan/Meituan-Robotics-0",
|
| 10 |
+
"code_commit": "88c667933dfc941a71c9b142eb84b716ac4f0707",
|
| 11 |
"serialization": "PyTorch state_dict; unchanged from the selected source checkpoint",
|
| 12 |
"initialization": "Stage 1 FAST checkpoint at 430000 training steps",
|
| 13 |
+
"repo_id": "MeituanRobotics/Meituan-Robotics-0",
|
| 14 |
+
"checkpoint_contains": [
|
| 15 |
+
"Qwen3.5-4B backbone",
|
| 16 |
+
"MMDiT action head"
|
| 17 |
+
],
|
| 18 |
+
"backbone_metadata": "base_vlm/Qwen3.5-4B-Action/",
|
| 19 |
+
"separate_qwen_weight_file_required": false,
|
| 20 |
+
"checkpoint_tensor_count": 929
|
| 21 |
}
|