arXiv:2609.18084v1 Announce Type: cross Abstract: Fine-tuning a Vision-Language-Action (VLA) model for a new deployment environment is expensive, yet most methods apply uniform-capacity adapters to e...
arXiv:2607. 23711v1 Announce Type: new Abstract: LoRA fine-tuning can create intruder dimensions: new leading singular vectors of the updated weight matrix $W+BA$ that are nearly orthogonal to all pretrained singular vectors and that drive catastrophic forgetting.
arXiv:2602. 16086v3 Announce Type: replace-cross Abstract: Recent collapse-free quantizers such as FSQ achieve stable training by replacing the learnable codebook with an engineered geometry: a fixed scalar grid whose structure is dictated by the codebook size K.
arXiv:2608.28706v1 Announce Type: new Abstract: ViT detectors fix a uniform token grid before any learned stage. A native-resolution aerial detector must then choose between resolving few-pixel objec...
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.
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.
The paper introduces a training‑free, human‑in‑the‑loop anomaly detection framework that allows a domain expert to correct a PatchCore detector by editing its memory bank, without retraining or using gradients. Using only ten golden samples, operator corrections close a median 66% of the performance gap to a fully trained bank, improving 12 of 15 MVTec AD categories while harming none. The approach is evaluated with a rigorous held‑out protocol and shows that passive and active querying yield statistically indistinguishable gains, with a defect‑memory extension failing decisively.
The paper investigates post‑training quantization of transformer attention blocks by optimizing a joint loss over the Q, K, V projections rather than individual weight matrices. Using this joint attention‑based objective (JAB), the authors achieve significant compression on Mistral‑7B, recovering 77‑90% of the performance gap at 3 bits, but the method fails when MLP layers are included. A role‑aware offset rule that ignores sensitivity estimates outperforms JAB on GPT‑2 and full Mistral‑7B, demonstrating that the matrix a weight belongs to is more critical than sensitivity metrics.
arXiv:2608.08477v5 Announce Type: replace Abstract: We build VectraYX-Vision-1B, a sub-2B Spanish/LATAM cybersecurity vision-language model for offline use, and measure what limits its visual groundi...
arXiv:2609.39223v2 Announce Type: new Abstract: Large language model (LLM) inference is increasingly moving toward lower precision to realize the throughput of hardware accelerators, but aggressive p...
arXiv:2608.23816v1 Announce Type: new Abstract: Quantized fine-tuning (QLoRA) saves memory but not time. It dequantizes every 4-bit weight on the fly, so it trains more slowly than fp16 LoRA. We pres...
arXiv:2606. 20128v1 Announce Type: cross Abstract: Benchmarks for LLM-generated GPU kernels (KernelBench, TritonBench, GEAK) score correctness through fixed-shape, small-sample allclose-style checks.