block-influence

Here is 1 public repository matching this topic...

VedantPandhare / Language-Balanced-Depth-Pruning-of-Small-Multilingual-LLMs

Investigates whether transformer layer redundancy in small multilingual LLMs (Qwen, LLaMA, Gemma) is language-dependent. Tests if English-calibrated layer pruning silently harms Hindi and Marathi performance, and whether a mixed calibration set fixes it.

transformers multilingual-nlp layer-pruning llm block-influence

Updated Oct 16, 2026
Jupyter Notebook

Add this topic to your repo

To associate your repository with the block-influence topic, visit your repo's landing page and select "manage topics."

Learn more