PolyLM fine-tunes a 9B-parameter LLM on 185k papers to predict polymer properties from text alone, achieving median R² of 0.74 on 68k held-out samples.
Unsupervised word embeddings capture latent knowledge from materials science literature.Nature, 571(7763):95–98
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Can LLMs Predict Polymer Physics Just by Reading Synthesis and Processing Prose?
PolyLM fine-tunes a 9B-parameter LLM on 185k papers to predict polymer properties from text alone, achieving median R² of 0.74 on 68k held-out samples.