DSpark draft model for Muse Spark Open proposes 7 tokens per pass with 23-42% acceptance
vllm_project · x · 2026-10-16
Inferact trained a DSpark speculative-decoding draft model for Muse Spark Open: block diffusion proposes up to 7 draft tokens per pass, confidence-based adaptive verification schedules checks, and sliding-window attention keeps drafting light at long context. Recipe supports CUDA and ROCm. Measured acceptance: GSM8K 42.04% (3.94 tokens/cycle), HumanEval 37.23%, SWE-bench Pro 23.69%. The 5-layer draft taps target hidden states at layers 13/29/45/57/61 and weighs 10.6GB.
Related event: vLLM and Inferact Optimize Muse Spark Open Serving(2 posts)→
More from Infra
- Berkeley scholars: concentrated chip supply chain makes a hardwired frontier AI pause feasible — wfithian · 2026-10-16
- Microsoft trained MAI-Base-2 on 16k GB200 GPUs in just 38 days, with 21% MFU — scaling01 · 2026-10-16
- Robotics data tokenization landscape: Khala maps peaq, THOR, BitRobot and Axis — 0xSammy · 2026-10-16
- Signal65 benchmarks: Intel Xeon 6 runs agent workloads 11% faster for less money — ryanshrout · 2026-10-16
- Microsoft's MAI-Base-2: 2.8T-param MoE pretrained on 16k GB200s in just 38 days — scaling01 · 2026-10-16
- Cloudflare and OpenAI pilot: real-time network signals bring unseen pages into ChatGPT search — irvinebroque · 2026-10-16