Long-context LLM serving is increasingly bottlenecked by KV-cache memory, especially in resource-constrained scenarios. Among existing KV-cache compression strategies, token-wise methods reduce cached states but risk information loss through eviction or condensation, while feature-wise methods reduce per-token KV dimensions but can require full-dimensional reconstruction to apply positional embedding, limiting decoding speedups. We introduce SlimKV, a question-agnostic joint token-feature KV-cache compression method. SlimKV uses low-rank-aware training to compress long contexts into beacon memory states with latent KV representations, together with layer-adaptive rank allocation. We further uncover a positional asymmetry: removing key-side RoPE affects beacon and raw tokens differently, with much smaller degradation for beacon tokens. Exploiting this asymmetry, SlimKV trains beacon KV projections under a K-RoPE-free constraint and enables latent-space attention during decoding, mitigating reconstruction latency. On LongBench, SlimKV outperforms baselines at 16×/32× compression and remains leading at 4×/8×, where it retains over 96% of the uncompressed model’s score. Needle-in-a-Haystack confirms robustness across evidence positions, and efficiency evaluation shows up to 7.34× attention speedup and 3.38× end-to-end decoding speedup over the uncompressed model at 128K length.

Large language models (LLMs) demonstrate strong general language capabilities but remain limited in chemical reasoning, particularly for tasks requiring structured, mechanistic understanding of molecular reactions. We present Knowledge Graph Reaction LLM (KGRxn-LLM), a framework that augments LLMs with a hierarchical chemical knowledge graph (KG) to ground reasoning in molecular transformations and reaction patterns. Existing benchmarks primarily emphasize reaction or molecular fact recall, providing limited assessment of reaction-level mechanistic reasoning. To address this gap, we introduce KGRxn-Bench, a benchmark of 1,200 questions designed to evaluate LLMs on reaction-centric reasoning tasks, including functional group identification, reaction type classification, and product and reagent prediction. Experimental results show that our approach of grounding LLMs in structured KG substantially improves performance across multiple tasks and model backbones, outperforming domain-specific fine-tuned models on KG-covered splits and most hold-out splits.