DSpark
A speculative decoding algorithm released by DeepSeek, engineered specifically for the Multi-head Latent Attention (MLA) architecture with a dedicated draft strategy. It has been merged into the vLLM main branch and supports the Qwen3 and Gemma model series.
Timeline (last 90 days)
DeepInfer has open-sourced core components such as the optimized inference engine, kernels, expert parallelism, and DSpark speculative decoding, but the distributed KV cache system, the request load balancer, and the coordination infrastructure across prefill/decode replicas are not included.
Across three target model sizes and two temperature settings, TraceSpec improves replay throughput by 12.5% to 22.6% compared with its DSpark baseline.
A DSpark configuration using 4 R9700-31K graphics cards reaches 55–69 tok/s running DeepSeek V4.1 Flash under Strata 0.2.4.
DSpark in DeepInfer first estimates the probability that a draft will pass, then, combined with hardware throughput, adjusts how long a draft each request is worth verifying.
DSpark is a speculative decoding implementation that works in concert with LES's expert preloading to further amplify on-device throughput gains.
The implementation, including expert scheduling and DSpark integration, has been open-sourced in llama.cpp-omni.
On the same benchmark with the Qwen3.8-27B model, DFlash2 achieves a 2.80x speedup, DSpark 2.57x, and MTP 2.22x.
DLite outperforms the current state-of-the-art method DSpark by up to 70% under agentic workloads.
Muse Spark Open includes a mechanism called DSpark, whose specific meaning is not explained in detail in the original information.
Meta provides two dedicated draft models for speculative decoding: DFlash V2 is designed for SGLang and is reportedly achieving a 3.7x throughput improvement on AMD MI355X hardware, accepting an average of 6.4 draft tokens per step; DSpark is designed for vLLM and proposes up to seven tokens per pass.
10 more timeline events
All Facts (19)
DeepInfer has open-sourced core components such as the optimized inference engine, kernels, expert parallelism, and DSpark speculative decoding, but the distributed KV cache system, the request load balancer, and the coordination infrastructure across prefill/decode replicas are not included.
50%UnverifiedAcross three target model sizes and two temperature settings, TraceSpec improves replay throughput by 12.5% to 22.6% compared with its DSpark baseline.
50%UnverifiedA DSpark configuration using 4 R9700-31K graphics cards reaches 55–69 tok/s running DeepSeek V4.1 Flash under Strata 0.2.4.
50%UnverifiedDSpark in DeepInfer first estimates the probability that a draft will pass, then, combined with hardware throughput, adjusts how long a draft each request is worth verifying.
50%UnverifiedDSpark is a speculative decoding implementation that works in concert with LES's expert preloading to further amplify on-device throughput gains.
50%UnverifiedDLite outperforms the current state-of-the-art method DSpark by up to 70% under agentic workloads.
50%UnverifiedMuse Spark Open includes a mechanism called DSpark, whose specific meaning is not explained in detail in the original information.
50%UnverifiedMeta provides two dedicated draft models for speculative decoding: DFlash V2 is designed for SGLang and is reportedly achieving a 3.7x throughput improvement on AMD MI355X hardware, accepting an average of 6.4 draft tokens per step; DSpark is designed for vLLM and proposes up to seven tokens per pass.
50%UnverifiedDSpark is a representative baseline method in the field of speculative decoding, belonging to one of the routes of joint optimization of the draft and the target model, and is used as a reference in JetSpec++'s comparison as the current state-of-the-art (SOTA) method.
50%UnverifiedJetSpec++ achieves a 1.54x wall-clock time improvement relative to DSpark.
50%UnverifiedCompared with the DSpark baseline, ReaDy improves on all benchmarks, with throughput increasing by an average of 7.97% and accepted length increasing by an average of 8.50%.
50%UnverifiedReaDy is built on top of DSpark's parallel backbone and sequence prediction head, using lightweight two-stage adaptation.
50%UnverifiedMiniCPM-V 4.7 35B-A3B uses a combination of three techniques—Lookahead Expert Scheduling (LES), fine-grained layer-by-layer quantization, and DSpark speculative sampling—to accelerate on-device inference.
50%UnverifiedLFM2-VL achieves a 1.66x to 2.56x speedup at batch 1 via DSpark speculative decoding.
50%UnverifiedDFlash and DSpark have made targeted optimizations in draft generation efficiency and hardware adaptation, and may outperform classic solutions under specific models and workloads.
50%UnverifiedvLLM v0.28.0 introduces DFlash2 and DSpark confidence-scheduled verification in the speculative decoding domain.
50%UnverifiedThe Q8 main model occupies 162 GB, preventing the drafter from fitting into unified memory, rendering DSpark acceleration ineffective; after moving three layers of MoE experts back to the CPU, output only improved from 22.87 to 23.49 tokens/second (2.7% faster), while prefill dropped from 95.38 to 76.65.
50%UnverifiedDeepSeek's official DSpark is essentially speculative decoding/multi-token prediction technology, requiring a drafter model and two sets of KV Cache resident in high-speed memory to achieve optimal results.
50%UnverifiedThe implementation, including expert scheduling and DSpark integration, has been open-sourced in llama.cpp-omni.
50%