DroidSpeak: KV Cache Sharing for Cross-LLM Communication and Multi-LLM Serving
read the original abstract
Compound AI systems, such as agentic systems, are an emerging trend in large-scale enterprise settings, with multiple LLMs specialized for different users, tasks, and/or roles working together. In these scenarios, different models often process inputs that share the same context prefix. Although much work was done in the past to enable the reuse of prefix KV caches across inputs for a single model, how to enable one model to reuse the prefix KV caches of a different model remains an open question. We introduce DroidSpeak, the first distributed LLM inference system that enables KV cache reuse across distributed nodes running inference of different LLMs, so long as the LLMs have the same architecture. We present the first study that aims at understanding the impact of sharing KV caches across different LLMs, and if/when such sharing affects quality. Inspired by the findings, we present DroidSpeak, which selectively recomputes a few layers of the KV cache produced by another LLM and reuses the remaining layers, with negligible quality loss. Moreover, carefully pipelining the layer-wise re-computation and the loading of reused KV cache further improves the inference performance. Experiments on diverse datasets and model pairs demonstrate that DroidSpeak achieves up to 4x throughput improvement and about 3.1x faster prefill (time to first token), with negligible loss of quality in F1 scores, Rouge-L or code similarity score, compared to the baseline which does not allow any sharing across models.
This paper has not been read by Pith yet.
Forward citations
Cited by 8 Pith papers
-
See What I See, Know What I Think: Dense Latent Communication Across Heterogeneous Agents
Heterogeneous agents achieve dense latent KV-cache communication via lightweight cross-model transformation and two-phase training, outperforming text at lower compute in context-aware settings and enabling context-un...
-
A Policy-Driven Runtime Layer for Agentic LLM Serving
Introduces a three-tier architecture with an agent runtime layer and four primitives for agent-aware policies in LLM serving, validated on KV caching via CacheSage showing 13-37pp hit-rate gains on five workloads.
-
Kamera: Unified Position-Invariant Multimodal KV Cache for Training-Free Reuse
Kamera stores a low-rank patch with each position-free KV chunk to restore cross-chunk conditioning lost in naive reuse, enabling cheap reordering, sliding windows, and recall across attention mechanisms.
-
PEEK: Context Map as an Orientation Cache for Long-Context LLM Agents
PEEK maintains a constant-sized context map via a programmable cache policy to give LLM agents persistent orientation knowledge about recurring external contexts, yielding 6-34% gains and lower cost than prior prompt-...
-
ForkKV: Scaling Multi-LoRA Agent Serving via Copy-on-Write Disaggregated KV Cache
ForkKV uses copy-on-write disaggregated KV cache with DualRadixTree and ResidualAttention kernels to deliver up to 3x throughput over prior multi-LoRA serving systems with negligible quality loss.
-
TokenDance: Scaling Multi-Agent LLM Serving via Collective KV Cache Sharing
TokenDance scales multi-agent LLM serving to 2.7x more concurrent agents by collective KV cache reuse and block-sparse diff encoding that achieves 11-17x compression.
-
OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelism
OxyGen unifies KV cache management in MoT VLAs to enable cross-task KV sharing and cross-frame continuous batching, delivering up to 3.7x speedup with 200+ tokens/s language and 70 Hz action on on-device platforms.
-
Semantic Cache Distillation: Efficient State Transfer via Reuse and Selective Patching
SCD replaces raw KV cache transmission with compact semantic codes via reuse and patching to achieve up to 2.65x TTFT speedup while staying within 5% F1 of oracle quality.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.