The authors define an interaction state as the task instruction plus the sequence of executed actions and resulting observations, excluding the student's earlier reasoning traces.
Grounding distillation in interaction states improves multi-turn agent performance
State-OPD separates executed interaction history from student reasoning, improving teacher supervision and broadening the actions sampled during training.
Chinese Tech
Zhentao Zhang · Yanzhao Zheng · Yanlin Liu · Yuanqiang Yu · Tianze Xu · Chao Ma · +6 more
Alibaba Group · Tsinghua University · Shanghai Jiao Tong University · University of Science and Technology of China
Research Digest··3 min read
Zhang et al.
Why this paper
From Alibaba Group and 3 others
In one line
In multi-turn agent distillation, grounding supervision in interaction states and resampling complete reasoning-action responses improves task success and action diversity.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 4B)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§