MOAI speeds non-interactive homomorphic transformer inference
It cuts BERT-base evaluation time 52.8% versus THOR and reports 2.36 minutes amortized per input on one GPU.
Stories per day · 30 days
3 RSS
Relatedqwen8gemma5inference3mistral3anthropic2botnet2iot2kv-cache2self-replication2svd2ai-safety1ai-training1ai-worms1bert1complement-consistency1compression1continual-learning1copyright1
45 stories, newest first · one card per story
Stories tagged #llama
Study tests whether causal neuron groups are unique or transferable
A few neurons often flip two-hop answers, but effects weaken on other chains.
Persona generalization follows a shared line across Qwen and Llama
Fine-tuning only later layers keeps strong in-domain learning while reducing out-of-domain unsafe transfer.
MoE Routing Strategies: Top-K, Expert-Choice, Sigmoid Gating and Dynamic Routing Explained
The guide covers seven production MoE routing methods used in models like Mixtral and DeepSeek-V4, highlighting sigmoid gating and bias balancing as dominant approaches.
1
LoRA merge audit finds bad adapters inflate forgetting scores
Naive merging of an audited four-adapter set cut GSM8K by 10.5 points and HumanEval by 18.3.
RidgeFill speeds prefill with a training-free late-layer cache
A closed-form ridge map fills upper-layer cache; reported TTFT gains are 1.25–1.60x on tested Qwen setups.
Skill-Uni jointly trains LLM skill retrieval and use
Authors report average gains of 66.59% on Qwen3-4B and 34.82% on LLaMA3.1-8B across five benchmarks.
LEMS and KFAC-SVD improve SVD compression of large language models
Zero-shot gains reach 4.8 points through 70B models, and KFAC-SVD lowers perplexity 15% on average.
Small LLMs fail complement consistency in agent opinion simulations
Paired true and false reports from Qwen3-4B, Llama-3.2-3B and Gemma-3-4B do not sum to one, and complementary populations diverge within five rounds.
LEMS advances SVD compression for LLMs
LEMS and KFAC-SVD deliver up to 4.8 percentage point accuracy gains on Mistral, Qwen3 and Llama 3 models.
Adaptive multi-token prediction selects draft horizons online
On Llama and Qwen pairings, it was up to 4.1x faster than standard decoding and beat a fixed window in 8 of 10 settings.
Court orders Meta to name executives who approved AI training torrenting
The order also covers who approved downloading shadow libraries for Llama 1 and 2 and what happened to unused pirated works.
FuTransformer speeds Llama-style inference with FP8 prefill and INT4 decode
On one H100, prefill reaches 1.29 PFLOP/s and is up to 1.48x faster than compiled PyTorch FP8.
Anthropic warns self-replicating AI could spread to your smart devices
Anthropic warns small models may soon self-replicate across insecure IoT devices, though its lab test did not produce a full worm.
SAE-IDF matches symbols across systems using sparse autoencoder codes
On 12 tasks across six symbolic systems, SAE-IDF beats the same LLM's dense baseline by up to 45 points.
Anthropic conjures another threat that doesn't exist yet to stoke fears of AI apocalypse
The Register says Anthropic's self-replicating AI worm research overstates a threat that does not yet exist.
Rethinking LLM-Guided Visual Question Answering: Does Deeper Knowledge Integration Really Help?
Study finds shallow LLM guidance beats deep fusion for knowledge-based VQA under scarce data.
Byzantine-Resilient Federated Large Language Model via Hypergraph Message Passing
HMP-GAE filters Byzantine updates during federated training of Qwen-2.5 and Llama-3.2.
Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State Quantization
ZIP-SR and ZE-EDEN cut 4-bit AdamW's validation-loss gap to 32-bit AdamW by up to 70%.
Poster: A Preliminary Study of LLM Distillation Inference
Distillation-inference test flags Llama-3.2-3B models distilled from Qwen2.5-7B at TPR 1.0 and p=0.02.
SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inference
SparseDecoding prunes LLMs using decoding activations, speeding generation up to 1.48x.
SLDR: Defending Against Malicious Fine-tuning via Selective Layers Recovery and Dynamic Routing
SLDR restores LLM refusal after malicious fine-tuning by recovering sensitive layers and routing only harmful queries.
Newer and Bigger, but Safer? A Longitudinal Study of the Functionality-Security Gap in LLM-Generated Code
Across 32 LLMs, newer models are safer in absolute terms, but no family closes the code security gap.
CARE: Certifying Acceleration for Vision-Language-Action Inference
CARE certifies VLA inference accelerators so speedups stay inside a user-specified failure budget.