How to use from the
Use from the
Transformers library
# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("XiangyunHuang/Route2Step", device_map="auto")
Quick Links

Route2Step

This repository contains the released Route2Step checkpoints.

Qwen3-VL-4B release

MAG-Qwen3-vl-4b/ is the Qwen3-VL-4B MAG checkpoint for local action execution. In our reported evaluation, it reaches an SR of 62.

The folder contains the inference-required model shards, configuration, tokenizer, and image/video preprocessing files. Training-only files are omitted.

Available checkpoints

  • MIA/: released MIA checkpoint.
  • MAG/: released MAG checkpoint.
  • MAG-Qwen3-vl-4b/: Qwen3-VL-4B MAG release.

Load the Qwen3-VL-4B checkpoint with a recent transformers version supporting Qwen3-VL:

from transformers import AutoProcessor, Qwen3VLForConditionalGeneration

model_id = "XiangyunHuang/Route2Step"
model = Qwen3VLForConditionalGeneration.from_pretrained(
    model_id,
    subfolder="MAG-Qwen3-vl-4b",
    torch_dtype="auto",
    device_map="auto",
)
processor = AutoProcessor.from_pretrained(model_id, subfolder="MAG-Qwen3-vl-4b")
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support