logits commited on
Commit
e9bb949
·
verified ·
1 Parent(s): 0cf04dd

Convert to the alias-free PI0.5 layout (813 tensors)

Browse files
Files changed (2) hide show
  1. README.md +19 -11
  2. model.safetensors +2 -2
README.md CHANGED
@@ -9,36 +9,44 @@ base_model:
9
  ---
10
  # FlashVLA · π0.5 · RoboTwin 2.0
11
 
12
-
13
  A **π0.5** flow-matching vision-language-action policy finetuned on **RoboTwin 2.0**
14
  (50-task multitask) and served with [**FlashVLA**](https://github.com/z-lab/flashvla)
15
  streaming action decoding for fast, asynchronous inference.
16
 
17
-
18
  - **Base model:** [`lerobot/pi05_base`](https://huggingface.co/lerobot/pi05_base)
19
  - **Method:** [FlashVLA](https://github.com/z-lab/flashvla) — streaming action decoding for flow-matching VLAs (async chunk-overlap execution)
20
  - **Benchmark:** RoboTwin 2.0, 50-task multitask (clean / randomized)
21
  - **License:** Apache-2.0
22
 
23
-
24
  ## Results
25
 
26
-
27
  RoboTwin 2.0 50-task multitask success rate (%). `d` is the async step delay:
28
  `d=0` is synchronous, `d=1`/`d=2` overlap the next chunk's inference with execution.
29
 
30
-
31
  | Model | Clean | Random | Avg |
32
  |:--|:--:|:--:|:--:|
33
- | π0.5 (base) | 86.1 | 85.8 | 86.0 |
34
- | **+FlashVLA** (`d=0`) | 90.8 | 90.2 | 90.5 |
35
- | **+FlashVLA** (`d=1`) | **91.2** | 89.9 | **90.6** |
36
- | **+FlashVLA** (`d=2`) | 91.0 | **90.2** | **90.6** |
37
 
38
 
 
39
 
 
40
 
41
- ## Usage
 
 
 
 
 
 
 
 
 
 
42
 
 
43
 
44
- Install the [FlashVLA](https://github.com/z-lab/flashvla) library.
 
9
  ---
10
  # FlashVLA · π0.5 · RoboTwin 2.0
11
 
 
12
  A **π0.5** flow-matching vision-language-action policy finetuned on **RoboTwin 2.0**
13
  (50-task multitask) and served with [**FlashVLA**](https://github.com/z-lab/flashvla)
14
  streaming action decoding for fast, asynchronous inference.
15
 
 
16
  - **Base model:** [`lerobot/pi05_base`](https://huggingface.co/lerobot/pi05_base)
17
  - **Method:** [FlashVLA](https://github.com/z-lab/flashvla) — streaming action decoding for flow-matching VLAs (async chunk-overlap execution)
18
  - **Benchmark:** RoboTwin 2.0, 50-task multitask (clean / randomized)
19
  - **License:** Apache-2.0
20
 
 
21
  ## Results
22
 
 
23
  RoboTwin 2.0 50-task multitask success rate (%). `d` is the async step delay:
24
  `d=0` is synchronous, `d=1`/`d=2` overlap the next chunk's inference with execution.
25
 
 
26
  | Model | Clean | Random | Avg |
27
  |:--|:--:|:--:|:--:|
28
+ | π0.5 (base) | 82.74 | 76.76 | 79.75 |
29
+ | **+FlashVLA** (`d=0`) | 90.64 | 90.06 | 90.35 |
30
+ | **+FlashVLA** (`d=1`) | **91.14** | **90.60** | **90.87** |
31
+ | **+FlashVLA** (`d=2`) | 90.20 | 89.66 | 89.93 |
32
 
33
 
34
+ ## Usage
35
 
36
+ Install the [FlashVLA](https://github.com/z-lab/flashvla) library.
37
 
38
+ See [`sim_eval/robotwin/`](https://github.com/z-lab/flashvla/tree/main/sim_eval/robotwin) for evaluation setup.
39
+
40
+ ## Citation
41
+
42
+ ```bibtex
43
+ @article{flashvla2026,
44
+ title = {{FlashVLA: Streaming Action Decoding for Fast and Asynchronous VLA Inference}},
45
+ author = {Li, Zekai and Tang, Jiaming and Liu, Zhijian},
46
+ year = {2026}
47
+ }
48
+ ```
49
 
50
+ ## License
51
 
52
+ Apache-2.0
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1d21a2b83800f8836de25783d542ed1dc9912a9c5b514a21f42bfdaf65b36893
3
- size 14467145992
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:143af9b4890dedad78bb1246dd144069f7fdb50872e964de2421bb88a455730e
3
+ size 16573735272