TorchSpec open-sources 3 SOTA speculative decoding draft models for Kimi K3
hongyangzh · x · 2026-09-04
The TorchSpec team has released the Kimi K3 Draft Collection: three SOTA speculative decoding draft models (EAGLE-3, DFlash2, DSpark) trained with vLLM on NVIDIA GB200, with data recipes published. They also note a key empirical finding: high acceptance rate doesn't equal faster decoding — each draft architecture has its own drawbacks.
Related event: Three Open-Source Draft Models for Kimi K3 Released(3 posts)→
More from Infra
- Corporate token spending growth collapses from 42% to 0.2% as hyperscaler ROIC falls to 20% — SumitGup · 2026-10-21
- Qualcomm plans worldwide performance-based layoffs as it chases $5B AI data center revenue — rohanpaul_ai · 2026-10-21
- vLLM at PyTorchCon 2026: KV cache and kernel optimizations across TPU, AMD, and NVIDIA — PyTorch · 2026-10-21
- antirez to port DeepSeek Infer kernels to DGX Spark's SM121 for DwarfStar — antirez · 2026-10-21
- Ternary-Bonsai-2-9B: 2-bit ternary vision-language model trends on Hugging Face for on-device use — prism-ml · 2026-10-21
- General Compute x Positron: GPUs Handle Prefill, Memory-First Chips Tackle Decode-Bound LLM Inference — cneuralnetwork · 2026-10-21