Block-Level Weight-Space Structure Persists Under Post-Training: An Empirical Study Across LLM Families
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
The Flow has not summarised this story yet — read it at arXiv Machine Learning.
BellowsKV introduces a training‑free approach that maps the hidden states of multiple adjacent layers in a large language model to a single shared latent representation per token, allowing each layer to retrieve its own keys and values via a layer‑specific projection. The method retains the latent at full dimension and offers two compression strategies: BellowsKV‑Quant, which stores the latent at low precision, and BellowsKV‑Evict, which makes a shared eviction decision per block. Experiments on Llama‑3.1‑8B, Mistral‑7B, and Qwen2.5‑7B show that BellowsKV‑Quant preserves about 99% of full‑cache accuracy at 3.6× compression and 97‑98% at 13‑14×, outperforming existing cross‑layer methods that drop to at most 84% accuracy at 16× compression.
arXiv:2607. 14568v1 Announce Type: cross Abstract: A companion study ran a 35B mixture-of-experts model on a 2011 NVIDIA Tesla C2075 (Fermi, sm_20, 6GB) as a GPU-prefill/CPU-decode hybrid, because the 4-bit model did not fit in device memory (arXiv:2606.
DanLing NestedTensor is a PyTorch tensor abstraction that embeds multi‑ragged structure directly into the tensor, allowing packed values to carry partition information and logical dimension order. This design enables broadcasting, feature transformations, and reductions to automatically respect ragged axes while preserving the same representation through autograd and both eager and compiled execution. Benchmarks on an A100 show significant speedups—up to 3.39× over padding for BERT models and 2.40–4.32× for a Pairformer‑style workload—while dramatically reducing peak memory usage.
The paper introduces a system that enables first‑order LoRA training on phone NPUs by running all frozen‑weight multiplications on the NPU while keeping adapters and the optimizer on the CPU. It reuses weight tiles for input‑gradient computation, selects precision based on gradient sensitivity, and fuses remaining CPU operations to overlap with NPU execution. The approach is evaluated on Qualcomm Snapdragon devices with Qwen and Llama LLMs, demonstrating background training for nightly personalization and benchmark workloads.
arXiv:2605. 15250v3 Announce Type: replace-cross Abstract: Multi-head Latent Attention (MLA), the attention used in DeepSeek-V2/V3, jointly compresses keys and values into a low-rank latent and matches the H100 roofline almost perfectly.
The paper investigates whether training Mixture-of-Experts (MoE) routers can improve memory‑bandwidth locality on consumer GPUs. Using a new zero‑surgery telemetry tool, the authors measure that a large Qwen3‑235B model is bottlenecked by disk‑based expert access, and that an LRU cache can serve a majority of requests. They pre‑register experiments training 137 M‑parameter MoE models with locality‑aware losses, finding that while cache misses can drop up to 60 % (99 % static‑pin hit rate), every configuration fails to meet a strict 1 % perplexity threshold, indicating a tight coupling between cache efficiency and model quality.