5-7B-Instruct agents on ALFWorld, which tests household tasks, and WebShop, which tests multi-step product search and purchasing.
Context-aware credit assignment improves language agents on long-horizon tasks
CCPO uses context-weighted trajectory comparisons and short-horizon progress to identify which actions contributed to eventual success.
Big Tech
Huiwen Han · Ziyao Zeng · Zihao Zhu · Sisong Bei · Hao Wang · Hongyu Xu · +3 more
Stanford University · Texas A&M University · Yale University · Amazon · Meta AI
Research Digest··2 min read
Han et al.
Why this paper
From Amazon and 5 others
In one line
CCPO improves RL for LLM agents by weighting observation-matched peers by context and combining historical and future credit.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (3 noted)
- ✓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.
§