Contrastive-LM/CLM-v0.1-8B — new model trending #30 on Hugging Face
Contrastive-LM released CLM-v0.1-8B, a Qwen3-8B contrastive model for fast state-action scoring and agent verification.
Contrastive-LM released CLM-v0.1-8B, an Apache 2.0 System One model that scores candidate actions rather than generating text. Two projection heads on a frozen Qwen3-8B encoder were trained with bidirectional InfoNCE on about 60 million Nemotron Q&A pairs, 30 million synthetic hard negatives, and 1 million agentic trajectories. Zero-shot it matches Jev on computer-use, gaming, and tool use at up to 9× lower latency, and with about 1,000 candidates is 13× faster. Fine-tuned heads reach 81.6% on DeepSWE and 87.6% on Terminal-Bench 2.1; a multimodal CLM-35B is planned for early October.
- CLM-v0.1-8B adds state and action heads on frozen Qwen3-8B.
- Trained with bidirectional InfoNCE on about 91 million examples.
- Fine-tuned heads report 81.6% DeepSWE and 87.6% Terminal-Bench 2.1.
- Apache 2.0 weights score given candidates and do not generate text.
- A larger multimodal CLM-35B is planned for early October.
- Organizations
- Contrastive-LM
- AI models
- CLM-v0.1-8BQwen3-8B
From the original article
2 min read at huggingface.cotext-ranking · 434 downloads · 227 likes · license apache-2. 0
By Contrastive-LM · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.