· 3 items
arXiv ↗
Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text…
- What
- Paradee distills Kokoro-82M into an 8.07M parameter single-voice text-to-speech model.
- Why
- Provides an ultra-lightweight, high-quality local voice model ideal for fast agent integration.
- Impact
- Drops voice generation latency for agents
arXiv ↗
T-Search: An Open Agentic Retriever and Playground for Hard Multi-Ste…
- What
- T-Search launches as an open agentic retriever and playground for multi-step web searches.
- Why
- Offers an open-weight tool specifically engineered for bounded multi-round search workflows.
- Impact
- Improves autonomous web search loops
arXiv ↗
Behavior-Preserving KV Cache Compression
- What
- Behavior-Preserving KV Cache Compression proposes training-free eviction for long contexts.
- Why
- Optimizes memory overhead during long-form generation and agent multi-turn execution.
- Impact
- Saves memory in long-context inference