2 papers
cs.LG2026
When Can One Neuron Fix Repetition Loops in LLMs?
Aristotelis Lazaridis, Aman Sharma, Dylan Bates +3
The Gemma 4 instruction-tuned models share a reproducible failure: on long factual enumeration prompts, such as TV episodes, the 88 IAU constellations, or the 151 original Pokemon,…
cs.LG2025
DTRNet: Dynamic Token Routing Network to Reduce Quadratic Costs in Transformers
Aman Sharma, Saeed Najafi, Parsa Farinneya +6
Transformers achieve state-of-the-art results across many tasks, but their uniform application of quadratic self-attention to every token at every layer makes them computationally…