Abstract
Production LLM systems often rely on separate models for safety and other classification-heavy steps, increasing latency, VRAM footprint, and operational complexity. We instead reuse computation already paid for by the serving LLM: we train lightweight probes on its hidden states and predict labels in the same forward pass used for generation. We frame classification as representation selection over the full token×layer hidden-state tensor, rather than committing to a fixed token or fixed layer (e.g., first-token logits or final-layer pooling). To implement this, we introduce a two-stage aggregator that (i) summarizes tokens within each layer and (ii) aggregates across layer summaries to form a single representation for classification. We instantiate this template with direct pooling, a 100K-parameter scoring-attention gate, and a downcast multi-head self-attention (MHA) probe with up to 35M trainable parameters. Across safety and sentiment benchmarks our probes improve over logit-only reuse (e.g., MULI) and are competitive with substantially larger task-specific baselines, while preserving near-serving latency and avoiding the VRAM and latency costs of a separate guard-model pipeline. Multi-backbone experiments on dense and mixture-of-experts architectures (Llama-3.2-3B, GPT-OSS-20B, Qwen3-30B-A3B) confirm that these findings generalize beyond a single model family.- Anthology ID:
- 2026.acl-long.1955
- Volume:
[Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](/volumes/2026.acl-long/)- Month:
- July
- Year:
- 2026
- Address:
- San Diego, California, United States
- Editors:
[Maria Liakata](/people/maria-liakata/),[Viviane P. Moreira](/people/viviane-p-moreira/unverified/),[Jiajun Zhang](/people/jiajun-zhang/unverified/),[David Jurgens](/people/david-jurgens/)- Venue:
[ACL](/venues/acl/)- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 42226–42239
- Language:
- URL:
[https://aclanthology.org/2026.acl-long.1955/](https://aclanthology.org/2026.acl-long.1955/)- DOI:
- Cite (ACL):
- Gonzalo Ariel Meyoyan and Luciano Del Corro. 2026. A BERTology View of LLM Orchestrations: Token- and Layer-Selective Probes for Efficient Single-Pass Classification. InProceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 42226–42239, San Diego, California, United States. Association for Computational Linguistics. - Cite (Informal):
[A BERTology View of LLM Orchestrations: Token- and Layer-Selective Probes for Efficient Single-Pass Classification](https://aclanthology.org/2026.acl-long.1955/)(Meyoyan & Del Corro, ACL 2026)- PDF:
[https://aclanthology.org/2026.acl-long.1955.pdf](https://aclanthology.org/2026.acl-long.1955.pdf)