Speculative decoding
Spiffy
Spiffy: Multiplying Diffusion LLM Acceleration via Lossless Speculative Decoding
On the frontierfirst seen Sep 22, 2025
Current frontier — recent, not yet superseded in the knowledge base
0 papers critique it · 0 beat it on benchmarks
What to use instead
Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.
- Collaborative Speculative Decoding (CoSpec)Beyond the Target: From Imitation to Collaboration in Speculative DecodingMay 24, 2026
- ToolSpecToolSpec: Accelerating Tool Calling via Schema-Aware and Retrieval-Augmented Speculative DecodingApr 15, 2026
- QuasarQuasar: Quantized Self-Speculative Acceleration for Rapid Inference via Memory-Efficient VerificationMar 2, 2026
- FLy (Training-Free Loosely Speculative Decoding)Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact MatchNov 28, 2025
- Nov 1, 2025
- Oct 30, 2025
- Oct 22, 2025
- Oct 8, 2025
- Group Tree Optimization (GTO)Bridging Draft Policy Misalignment: Group Tree Optimization for Speculative DecodingSep 26, 2025
- Sep 22, 2025