openjev — Qwen3.5 trained as jev model

openjev is Qwen3.5 turned into a jev model: a single cross-encoder that reads a premise and a hypothesis and answers with entailment, contradiction or neutral. That one primitive is enough to rerank answers, grade them against a reference, guard content, and play games in real time: hand it the game state and a few statements about it, and the argmax entailment is the move. Doom above is played zero-shot, first from the text state and then straight from the pixels through the Qwen3.5 vision tower. Nothing is trained per task.

What's inside

• qwen3.5-4b-nli/ — the 4B jev checkpoint (Qwen3_5ForSequenceClassification, 3 labels: contradiction, entailment, neutral, last-token pooling, trained with plain cross-entropy over the three classes).
• modeling_openjev.py — OpenJevCrossEncoder: predict, rerank, grade, latents.
• code/ — everything used here: the trainer, the multiple-choice harness, Flappy Bird and Doom (text and pixels), the radar.
• videos/ — Flappy Bird and Doom replays; results/ — raw JSON for every run and the full report.

Use it

from modeling_openjev import OpenJevCrossEncoder
jev = OpenJevCrossEncoder("AlexWortega/openjev", subfolder="qwen3.5-4b-nli")

jev.predict([("The bird is 0.05 below the centre of the gap.", "The bird is below the centre of the gap.")])
# -> [[contradiction, entailment, neutral]] probabilities

jev.rerank("Which gas do plants absorb during photosynthesis?", ["oxygen", "carbon dioxide", "nitrogen"])
# -> index of the option with the highest entailment

Or with plain transformers:

from transformers import AutoModelForSequenceClassification, AutoTokenizer
tok = AutoTokenizer.from_pretrained("AlexWortega/openjev", subfolder="qwen3.5-4b-nli")
model = AutoModelForSequenceClassification.from_pretrained("AlexWortega/openjev", subfolder="qwen3.5-4b-nli")
text = model.config.nli_template.format(premise="...", hypothesis="...")

Reference point: dleemiller's NLI cross-encoders. Licence MIT.