C3 method targets KV cache bottleneck in long-context speech language models
A new paper introduces C3, a speech-guided KV retrieval method that reduces the memory, bandwidth, and decoding costs of the KV cache during long-context SpeechLLM inference.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-03 05:00 · arXiv cs.LG
C3: Speech-Guided KV Retrieval for Long-Context SpeechLLM Inference