arXiv:2606. 02559v1 Announce Type: cross Abstract: Post-training compression of Large Language Models (LLMs) removes entire architectural components, either deleting them or replacing them with fitted modules.
arXiv:2609.40127v1 Announce Type: cross Abstract: Modern transformers pair impressive capabilities with substantial memory and compute demands. Low-rank weight factorization reduces both while keepin...
Modern transformers pair impressive capabilities with substantial memory and compute demands. Low-rank weight factorization reduces both while keeping the matrices dense, and thus efficient on standar...
SHIFT-LLM is a training‑free post‑pruning correction framework that inserts a Linear Residual Adapter (LRA) at each depth‑pruned site in large language models. Each LRA preserves the original residual identity while adding a lightweight affine correction calibrated via closed‑form least‑squares regression on a small held‑out set, thereby approximating the hidden state that would have been produced by the removed block. Experiments across multiple model families and benchmarks show that SHIFT‑LLM consistently recovers accuracy lost to depth pruning, achieving gains up to +15.7 points on Llama‑3.1‑8B‑Instruct with only a few hundred calibration samples and no gradient computation.
arXiv:2607. 18363v1 Announce Type: cross Abstract: Feed-forward networks hold two thirds of a transformer's non-embedding parameters, yet the architecture has not received a necessity test that controls parameters, compute, and depth at once.
Task-Aware Spectral Pruning (TASP) is a post‑training framework that tailors sparse masks to specific tasks by calibrating module‑level spectral descriptors against task‑specific ablation effects. It constructs masks that close grouped‑query‑attention and SwiGLU dependencies, routing each user turn to a single compiled mask that remains fixed during prefill and decoding. In experiments, TASP achieves a 43% active‑FLOP reduction while preserving 97.7% of the dense BF16 performance on Llama‑3‑70B, and delivers a 1.44× speedup on an A100 80GB with INT8‑weight/BF16‑compute, reducing decode latency from 45.2 to 31.3 ms/token.
arXiv:2602.06471v2 Announce Type: replace Abstract: The architectural shape of dense Transformers has remained remarkably stable: narrow-wide-narrow feed-forward networks (FFNs) consume most non-embe...
XMerge is a post‑training method for compressing large language models by removing entire transformer layers while preserving a standard serving architecture. It selects low‑impact blocks via cross‑axis selection and refits adjacent surviving blocks with local boundary reconstruction, requiring no task labels or fine‑tuning. Across seven Llama and Qwen backbones, XMerge outperforms five published baselines, achieving top rankings on CORE and MMLU tasks even at aggressive compression levels, and consistently avoids model collapse while improving calibration and decoding efficiency.
arXiv:2607.12550v3 Announce Type: replace-cross Abstract: The key-value (KV) cache has become the dominant memory cost of transformer inference: it grows with batch size, context length, and depth, a...
arXiv:2607. 12550v1 Announce Type: new Abstract: The key-value (KV) cache has become the dominant memory cost of transformer inference.
The study re‑examines a reported advantage of a routed ternary (1.58‑bit) language model over a full‑precision transformer at 60K parameters. By running controlled experiments with multiple seeds and a fixed training recipe, the authors find that the apparent benefit largely stems from the choice of baseline model shape rather than the ternary architecture itself. While the routed model does outperform other shapes at a larger 130M‑byte budget, its advantage diminishes when a plain gated diagonal‑SSM block is used, and the ternary penalty varies with architecture and quantization details.
arXiv:2606. 27449v1 Announce Type: new Abstract: Multi-head attention conventionally partitions the hidden dimension equally across all heads at every layer, enforcing an identical representational subspace dimension (dh = dmodel/h) throughout the models depth.
arXiv:2602. 00161v2 Announce Type: replace-cross Abstract: In this paper, we formulate the compression of large language models (LLMs) by optimally deleting transformer blocks (``block removal'') as a constrained binary optimization (CBO) problem that can be mapped to a physical system (Ising glass), whose energies are a strong proxy for downstream model performance.