lm_head
PulseAugur coverage of lm_head — every cluster mentioning lm_head across labs, papers, and developer communities, ranked by signal.
-
Research Explains Why Post-Training Quantization Works for LLMs
A new research paper explores the phenomenon of post-training quantization (PTQ) in large language models (LLMs). PTQ compresses LLMs by reducing the precision of their weights, which typically introduces errors into th…
-
New speculative decoding methods boost LLM inference speed and safety
Researchers are developing advanced speculative decoding techniques to accelerate large language model inference. HyperDFlash optimizes decoding for DeepSeek-V4's multi-hyper-connection architecture, improving draft acc…
-
Google's Gemma 4 31B model gets style tune with minimal VRAM
A new fine-tuned version of Google's Gemma 4 31B model, named Gryphe/Gemma-4-31B-StyleTune, has been released. This fine-tuning process focused on altering only the lm_head tensor, which controls token emission, to sign…
-
LLM analysis method reveals training data secrets and ethical risks
Researchers have developed a method using singular value decomposition (SVD) of a large language model's weight matrix to reveal interpretable semantic subspaces. This technique, requiring minimal code and no model infe…