other

multi-turn language models

otheractiveprovisionalmulti-turn-language-models-95ed38d9·1 events·first seen 18d ago

Aliases: multi-turn language models

Co-occurring entities

on-policy distillation self-anchored drift Canonical-Context On-Policy Distillation (CCOPD)

More like this (12)

Transformer Language Models Reinforcement Learning for Language Models encoder-only language models Reasoning Language Models unsupervised language modeling Multimodal Large Language Models large language models generative language modeling large language model agents Large Language Models (frontier)Latent Context Language Models 1B-scale language models

Recent events (1)

6arXiv · cs.CL·18d ago·source ↗

Canonical-Context On-Policy Distillation (CCOPD) for Multi-Turn LLM Consistency

This paper identifies 'self-anchored drift' as a key failure mode in multi-turn LLMs: when information is revealed incrementally across turns, models produce unsupported assumptions that distort final answers, even when the total evidence is identical to a single-prompt setting. The authors propose Canonical-Context On-Policy Distillation (CCOPD), which trains a student model on incremental multi-turn conversations to match the output distribution of a frozen teacher conditioned on the full clean prompt. Trained only on math conversations, CCOPD achieves a 32% average relative improvement on multi-turn (RAW-SHARDED) tasks and generalizes zero-shot to five out-of-domain task families while preserving single-prompt performance.

Evaluation and Benchmarking Agent and Tool Ecosystem on-policy distillation multi-turn language models self-anchored drift +2 more