Instructions to use mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051", dtype="auto")

LeRobot
How to use mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051 with LeRobot:
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051

SGLang

How to use mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051 with Docker Model Runner:
```
docker model run hf.co/mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051
```

VLA-0 Gorgeous 30Hz Merged v2v3 Qwen3.5 2B

This repository contains a lightweight inference package exported from:

runs/vla0_gorgeous_30hz_merged_v2v3_qwen35_2b_8gpu

It is packaged for use with the VLA-0 codebase in an environment aligned to lerobot==0.5.1.

Included

model_last/
- Hugging Face save_pretrained export for the Qwen3.5-2B-based VLA-0 model.
config.yaml
- Training/inference configuration used by rv_train.train.get_pretrained_model.
dataset_stats.pkl
- Action de-normalization statistics required at inference time.
model_last.pth
- Lightweight compatibility stub for VLA-0 inference. The actual model weights live in model_last/model.safetensors.

Important note about `model_last.pth`

The original training checkpoint .pth file is very large because it also stores optimizer and scheduler state. For inference with VLA-0, the model weights are loaded from model_last/, while model_last.pth is only used as a compatibility entrypoint by the current loader.
This package therefore ships a compact inference stub instead of the original full training-state checkpoint.

Usage

Clone or download this repository locally, then point VLA-0 at the included stub:

ROBOVERSE_DEPLOY_CHECKPOINT=/path/to/this-repo/model_last.pth python rv_train/deploy/service.py

Or load the transformer export directly:

from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained("/path/to/this-repo/model_last")
processor = AutoProcessor.from_pretrained("/path/to/this-repo/model_last")

Compatibility

Base model: Qwen/Qwen3.5-2B
Intended environment: VLA-0 repository with lerobot==0.5.1
Export date: 2026-04-19

License

This package follows the licensing described in the upstream VLA-0 repository:

VLA-0 code and released checkpoints: CC BY-NC 4.0
Base model licensing follows Qwen/Qwen3.5-2B

Added numbered epoch checkpoint

model_52/ and model_52.pth were added on 2026-04-20.
This is the latest numbered epoch checkpoint currently available in the source run directory.
model_last is kept separately and is not identical to model_52 in the current run state.

Downloads last month: 1

Inference Providers NEW

Image-Text-to-Text

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for mimi-neoki/vla0-gorgeous-30hz-merged-v2v3-qwen35-2b-lerobot051

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B