Downloads · 30 days
12
14% of all-time downloads
Damirchik/vla-grounder-qwen3.5-9b-rl4vla-pi0
vla-grounder-qwen3.5-9b-rl4vla-pi0 is a image-text-to-text model from Damirchik. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers.
This checkpoint is the fully merged Qwen3.5-9B VLA Grounder trained on the RL4VLA MultiCarrot and MultiPlate tasks using sparse rollout rewards from a frozen pi0 policy. It receives a scene image and a human instructi…
Downloads · 30 days
12
14% of all-time downloads
All-time downloads
87
Public
Parameters
9.4B
18.8 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors18.8 GB · 100%
From the Hugging Face model README
This checkpoint is the fully merged Qwen3.5-9B VLA Grounder trained on the RL4VLA MultiCarrot and MultiPlate tasks using sparse rollout rewards from a frozen pi0 policy. It receives a scene image and a human instruction and generates a concise, visually grounded command for the downstream frozen VLA.
Qwen/Qwen3.5-9Bimport torch
from transformers import AutoModelForImageTextToText, AutoProcessor
model_id = "YOUR_HF_ORG/vla-grounder-qwen3.5-9b-rl4vla-pi0"
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
trust_remote_code=True,
)
Use the prompt construction and answer parsing implemented in the VLA Grounder repository. Only the final command extracted from the model response should be passed to the frozen VLA.
This checkpoint is intended for research on language-conditioned robot control, frozen-VLA adaptation, and reproducibility of the accompanying experiments.