Speed sweep

pg-01a026ab-646f-7a74-a325-7f4678968ae6-sweep

pg-01a026ab-646f-7a74-a325-7f4678968ae6-sweep

JSON API Back to speed sweep search

Record

Recipe: pg-gemma-4-12b-q4km-unknown-apple-m4-max-128gb-llama-cpp-68722cca18
Measured: 2026-08-21T23:32:41.699Z
Points: 4

Measured speed

Concurrency	Context	Prefill	Decode	TTFT ms	Status
1	8,256	—	46.9	—	observed
1	32,832	—	40.9	—	observed
1	262,080	—	16.5	—	observed
1	32,768	—	—	100,301.3	observed

Remaining fields

Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.

id: pg-01a026ab-646f-7a74-a325-7f4678968ae6-sweep
measured at: 2026-08-21T23:32:41.699Z
recipe id: pg-gemma-4-12b-q4km-unknown-apple-m4-max-128gb-llama-cpp-68722cca18
schema version: local-ai-registry/v1

metrics

base memory bytes: 12209291264
base memory context tokens: 192
concurrency: 1
decode32k context tokens: 32,832
decode32k tps: 40.94
decode8k context tokens: 8,256
decode8k tps: 46.914
decode max context tokens: 262,080
decode max context tps: 16.516
decode mode: non-mtp
inference engine version: 1 (32e41fa)
latest point at: 2026-08-21T23:32:41.699Z
max context tokens: 262,080
max prompt tokens: 262,016
memory8k bytes: 14460502016
memory8k context tokens: 8,256
memory max context bytes: 24293801984
memory max context tokens: 262,080
peak generation tps: 40.94
peak memory bytes: 24372871168
peak prompt tps: 326.696
point count: 224
ttft32k cached prompt tokens: 28,672
ttft32k context tokens: 32,768
ttft32k seconds: 100.301

rows

concurrency	context tokens	decode tok s	decode tok s per stream	output tokens	peak vram gb	prefill tok s	samples
1	8,256	46.914	46.914	Unknown	13.467	Unknown	1
1	32,832	40.94	40.94	Unknown	22.699	Unknown	1
1	262,080	16.516	16.516	Unknown	22.625	Unknown	1
1	32,768	Unknown	Unknown	Unknown	Unknown	Unknown	1

Provenance & metadata (1)

source

paths: publication:pg-20260827T060320709Z, run:01a026ab-646f-7a74-a325-7f4678968ae6
repository: local.ai ↗