Strix Halo 8060S DwarfStar ROCm 10.0 Qwen3.8-Flash-Next Q4 depth-0 decode

Protocol deterministic-prime-extend-completions-v1, runner local-llm-speed v1, measured at the OpenAI-completions API endpoint. Mean of 2 repeats after 1 warmup. Concurrency 1, so the figure is per-request, not aggregate. Depth is the primed KV prefix; at depth 0 the whole prompt is computed (actual cached depth 0, recomputed history 0). Prefill is effective follow-up prompt processing: uncached tokens divided by client-observed time to first output. Decode is client-observed generation throughput over the tokens after the first, so TTFT is excluded. KV cache type default, speculative decoding off. Source: local-llm-benchmarks.dev/data/results.json. Run run-8fd615a954deed7b, completed 2026-09-24. Engine DwarfStar 5efa5b1b0252ca02afe62e6b4b9d8ba25d5b6e06955f65272deff1f90f0b8fbd /usr/local/bin/ds4-server, runtime ROCm 10.0. Container docker.io/kyuz0/strix-halo-ds4-toolbox:rocm-10.0. Weights: local DwarfStar GGUF file Qwen3.8-Flash-Next-Q4.gguf, quant Q4. Prompt 2048 tokens, 128 generated, primed depth 0.

Baseline, speculative decoding off. The same dataset has a Native MTP run of this quant that is not recorded here.