format int64 | model string | title string | auto_title string | views list | params int64 | height_unit int64 | source dict | input dict | output dict | plots list | stages list | plates list | tokens int64 | unused list | not_drawn list | summary string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2 | auto/vla_adapter | VLA-Adapter (1.6B): published base weights, adapter untrained | vla_adapter (vla_adapter) | [
"wrote",
"changed"
] | 1,575,044,575 | 1,024 | {
"checkpoint": "weights.safetensors",
"step": null,
"package": "vla_adapter",
"code": "model.py",
"observation": null,
"hooked_vs_plain_max_difference": 0,
"traced_by": "rebuild.trace_any",
"checks": {
"camera grid": {
"changed cell": [
2,
12
],
"most moved token":... | {
"seed": 9000007,
"step": null,
"text": "Pick up the lime star",
"swap": false,
"pieces": [
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
"",
... | {
"poses": [
[
2.035,
3.4866,
1.607,
-1.5257,
-1.2687,
-1.4106,
0.0447
],
[
2.0177,
3.4892,
1.6032,
-1.4964,
-1.2707,
-1.4192,
0.0891
],
[
2.0005,
3.4918,
1.5995,
-1.4669,
-1.2728,
... | [
{
"id": "learned",
"label": "start token",
"kind": "learned",
"grid": [
1,
1
],
"tokens": [
0,
1
],
"at": [
54,
0
],
"auto": {
"label": "learned tokens"
}
},
{
"id": "cam_top",
"label": "top camera",
"kind": "image... | [
{
"id": "in.image",
"label": "camera image",
"kind": "input",
"group": [
"inputs"
],
"plots": [
"cam_top",
"cam_side",
"cam_iso"
],
"width": 3,
"module": "",
"params": 0,
"note": "512 x 512 pixels, red, green and blue, as the robot's cameras see th... | [
{
"id": "vla.base_model.model.language_model.model.layers.0.plate",
"stage": "vla.base_model.model.language_model.model.layers.0.self_attn",
"heads": 14,
"key": "plate/0",
"allowed": "plate/allowed"
},
{
"id": "vla.base_model.model.language_model.model.layers.1.plate",
"stage": "vla.... | 874 | [
{
"name": "vla.base_model.model.vision_backbone.featurizer.norm.weight",
"params": 1024
},
{
"name": "vla.base_model.model.vision_backbone.featurizer.norm.bias",
"params": 1024
},
{
"name": "vla.base_model.model.vision_backbone.fused_featurizer.norm.weight",
"params": 1152
},
{
... | [
{
"name": "vla.base_model.model.language_model.model.embed_tokens",
"class": "Embedding",
"params": 136134656,
"out": [
1,
106,
896
]
}
] | Answers the whole 8-step chunk in one pass, with no noise. Two vision encoders (DINOv2 and SigLIP) read each camera; a 0.5B language model (Qwen2.5) reads them with the instruction and 64 learned action queries; a head reads every one of its layers. Here the adapter is untrained. Its head splits what it reads at 512 im... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.