Qwen3.5-4B as a single NLI cross-encoder: from rerank to Doom

Alex Wortega (AlexWortega) released the openjev model on Hugging Face — Qwen3.5-4B, fine-tuned as an NLI cross-encoder Qwen3_5ForSequenceClassification with entailment/contradiction/neutral labels, applicable for rerank, reference-based evaluation, content moderation, and zero-shot Doom gameplay.

Compiled by Sergey KostenchukPublished 2026-09-16Updated 2026-09-18

2026-09-16 Research HuggingFace

🤖 Qwen3.5-4B as a single NLI cross-encoder: from rerank to Doom

Alex Wortega (AlexWortega) released the openjev model on Hugging Face: Qwen3.5-4B, fine-tuned with cross-entropy on latent representations of the base as a Qwen3_5ForSequenceClassification classifier with entailment, contradiction, and neutral labels. One model is enough for rerank, reference-based evaluation, moderation, and zero-shot Doom gameplay — from state descriptions and directly from pixels.

🌍 A single cross-encoder on a ready-made 4B model replaces a set of narrow head models — reranker, LLM response grader, moderation classifier — without per-task training: inference reduces to a single forward pass with argmax.

👤 The qwen3.5-4b-nli checkpoint is open under the MIT license: it can be run via transformers or the OpenJevCrossEncoder wrapper; training code, benchmarks, and results are in the repository.

Source 1: https://huggingface.co/AlexWortega/openjev Source 2: https://x.com/justALEXWORTEGA/status/2100341039986798930

Sources