The publication presents llama.cpp inference throughput benchmarks with MTP for Qwen 3.6 models on three GTX 1080 Ti GPUs. It says the tests used a Q4_0 K/V cache and draft-MTP flags.
The results include context sizes and token rates, but the available summary does not provide numerical values. The reported setup may help engineers compare inference on older GPUs.
To reproduce or compare the tests, record the model and version, flags, cache format, hardware, and context sizes. Compare rates under equivalent conditions; results can vary with software versions, workload, and configuration.
If you use AI to study or apply this material, avoid submitting internal prompts, data, or documents that are not necessary. Review your organization’s data-handling rules before sharing information with any service.