The authors conducted three complementary empirical analyses on personalized LLM adapters.
Personalized LLMs can share most adaptation capacity across users, needing only tiny per-user code.
LINEUP learns shared low-rank factors and composes them per query, reducing per-user parameters from 4.19 million to 8 scalars while outperforming baselines on all 12 metrics.
Big Tech
Songyuan Sui · Srikanth Malla · Chiho Choi · Joon Hee Choi
Samsung Semiconductor · Rice University
Research Digest··3 min read
The authors revisit personalized adaptation for large language models by asking how much of a personalized adapter is actually personal.
Why this paper
From Samsung Semiconductor and Rice University
In one line
Personalization of large language models can be achieved with shared low-rank factors and only eight scalars per user, replacing full private adapters.
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
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§