Models, datasets, and demos

Model weights

Release	Download
Valen 0.8B	Hugging Face
Valen 2B	Hugging Face
Valen 4B	Hugging Face
Valen Preview 0923	Earlier checkpoint

Explore the organization's model collection for available export formats.

Datasets

Dataset	Purpose
VisualDecisionBench	Image and video decision evaluation
JevalKit Datasets	Ready-to-use TSV views of eight benchmark families
General 100k	Earlier general training release
Valen Eval Game	Game training and evaluation data

Interactive resources

Source repositories

Valen · JevalKit

CODE EXAMPLE01

from transformers import AutoModel

model = AutoModel.from_pretrained(
    "Valen-Team/Valen-0.8B",
    trust_remote_code=True,
    dtype="auto",
    attn_implementation="sdpa",
).to("cuda").eval()

result = model.predict({
    "state": "A cat is sitting on the sofa.",
    "questions": {"animal": {
        "type": "choice",
        "instructions": "Which animal is present?",
        "criteria": {"cat": "A cat", "dog": "A dog"},
    }},
}, execution="shared_state")
print(result)

Load the model in Python with a compatible CUDA GPU and Transformers dependencies.

Open the full guide

Model weights​

Datasets​

Interactive resources​

Source repositories​

Model weights

Datasets

Interactive resources

Source repositories