Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs Paper • 2609.29845 • Published 18 days ago • 105
view article Article Transformers now runs llama.cpp quants +1 marcsun13, ArthurZ, lysandre • 20 days ago • 109
LLM Compression by Block Removal with Constrained Binary Optimization Paper • 2602.00161 • Published Jun 17 • 9
view article Article Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem MultiverseComputingCAI • 21 days ago • 34
Compile by Training: Turning Natural-Language Specifications into Local Neural Functions Paper • 2609.04199 • Published Sep 3 • 332
LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes Paper • 2609.03796 • Published Sep 3 • 182
view article Article Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps +1 iamleonie, burtenshaw, sergiopaniego • Sep 3 • 148
view article Article LFM2.5 Q4\_0 Checkpoints from Quantization-Aware Distillation LiquidAI • Aug 19 • 37
view reply Have you experimented with or tested different vocabulary sizes to see if the vocab size directly impacts the overtraining threshold and benchmark peak?