Update README.md
Browse files
README.md
CHANGED
|
@@ -1,9 +1,31 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
| 4 |
This repository provides the official release of the Pi 0.5 model with the stop-gradient mechanism enabled, trained with the whole VLABench's official primitive tasks dataset.
|
| 5 |
-
This checkpoint is trained on 8 H100 for 30k iterations, with 5000 episodes data acrossing 10 tasks.
|
| 6 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
The reference success rate of this model is:
|
| 8 |
| Track | add_condiment | insert_flower | select_book | select_chemistry_tube | select_drink | select_fruit | select_mahjong | select_painting | select_poker | select_toy | Avg_SR |
|
| 9 |
|----------------------------|---------------|---------------|-----------------|-----------------|--------------------|--------------------|-----------------|-----------------|-------------------|-------------------|--------|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
# Pi05 official implementation trained on VLABench datasets.
|
| 5 |
This repository provides the official release of the Pi 0.5 model with the stop-gradient mechanism enabled, trained with the whole VLABench's official primitive tasks dataset.
|
|
|
|
| 6 |
|
| 7 |
+
|
| 8 |
+
# Evaluation
|
| 9 |
+
To run this checkpoint, please clone this repo: https://github.com/Shiduo-zh/openpi, and checkout to the branch `pi05`.
|
| 10 |
+
Assume that you download this checkpoints and put it in the directory `checkpoints`, to run the policy as server, please run:
|
| 11 |
+
```sh
|
| 12 |
+
bash vla_bench_scipts/serve_policy.sh pi05_ft_vlabench_primitive checkpoints/VLABench/pi05-primitive-10task/sg_pi05_base/29999/
|
| 13 |
+
```
|
| 14 |
+
|
| 15 |
+
After serving the policy, open another terminal and run:
|
| 16 |
+
```sh
|
| 17 |
+
bash vla_bench_scipts/multi_run_vlabench.sh <Your path to store the evaluate results>
|
| 18 |
+
```
|
| 19 |
+
|
| 20 |
+
# Train
|
| 21 |
+
To reproduce the training result, please run the training script with the config `pi05_ft_vlabench_primitive`.
|
| 22 |
+
```sh
|
| 23 |
+
XLA_PYTHON_CLIENT_MEM_FRACTION=0.95 uv run scripts/train.py pi05_ft_vlabench_primitive --exp-name=pi05_ft_vlabench_primitive --overwrite
|
| 24 |
+
```
|
| 25 |
+
|
| 26 |
+
Our checkpoint is trained on 8 H100 for 30k iterations, with 5000 episodes data acrossing 10 tasks.
|
| 27 |
+
|
| 28 |
+
# Reference Results
|
| 29 |
The reference success rate of this model is:
|
| 30 |
| Track | add_condiment | insert_flower | select_book | select_chemistry_tube | select_drink | select_fruit | select_mahjong | select_painting | select_poker | select_toy | Avg_SR |
|
| 31 |
|----------------------------|---------------|---------------|-----------------|-----------------|--------------------|--------------------|-----------------|-----------------|-------------------|-------------------|--------|
|