R9700 ROCm 10.0 LFM2.5-1.2B-Instruct BF16 64k-depth 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-14fd9bfce147b605, completed 2026-09-01. Engine llama.cpp 10.0, runtime ROCm 10.0. Container r9700-llama-rocm-10.0. Weights: None file LFM2.5-1.2B-Instruct-BF16.gguf, quant BF16. Prompt 2048 tokens, 128 generated, primed depth 65536.

Decode falls from 184.23 tok/s at depth 0 to this figure at a 65536 primed prefix.