Ryzen AI 9 HX 370 (Radeon 890M) LocalScore: Llama 3.2 1B Instruct decode_tok_s
67.0 blended generation tok/s across LocalScore's 9 tasks
Fields
id
benchmarks/localscore-hx370-llama32-1b-q4km-genMetric
decode_tok_s
Value
67
Unit
blended generation tok/s across LocalScore's 9 tasks
Model
Llama 3.2 1B Instruct
Methodology
LocalScore public result for 'AMD Ryzen AI 9 HX 370 w/ Radeon 890M', 26.5 GB, https://www.localscore.ai/accelerator/721. Engine is llamafile (Llamafile wraps llama.cpp). Quant Q4_K_M. The published prompt tok/s, generation tok/s and TTFT are each a blend across nine tasks whose prompt lengths are 16, 64, 384, 1024, 1280, 2048 and 4096 tokens and whose generation lengths run from 16 to 3072 (see localscore.ai/about) — NOT a single (prompt, output) pair. Single device, per-request, community-submitted. status=draft because the blend cannot be unwound into one context length.
Measured by
third_party
Reproducible
yes
notes
Draft on purpose. Do not quote next to a llama-bench -p/-n figure; the prompt length is a nine-task blend. The iGPU (Radeon 890M) is not a separate accelerator record; the SoC id is the device.
Notes
Draft on purpose. Do not quote next to a llama-bench -p/-n figure; the prompt length is a nine-task blend. The iGPU (Radeon 890M) is not a separate accelerator record; the SoC id is the device.