BWhat changed
Copied a pretrained transformer, removed complementary top/bottom layer blocks and concatenated retained layers before continued training, increasing depth while reducing discontinuity at the join.
Scientific scope
Copied a pretrained transformer, removed complementary top/bottom layer blocks and concatenated retained layers before continued training, increasing depth while reducing discontinuity at the join. Layer reuse, model growth and continued pretraining have predecessors. Source uses Llama2 architecture initialized with Mistral7B weights; not a new backbone trained from scratch or MoE. A particular48-layer construction does not prove optimality or compatibility with all transformer implementations.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Layer reuse, model growth and continued pretraining have predecessors. Source uses Llama2 architecture initialized with Mistral7B weights; not a new backbone trained from scratch or MoE. A particular48-layer construction does not prove optimality or compatibility with all transformer implementations.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling
- Kim et al.: pretrained-layer depth up-scaling — author/institutional project
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction