Speed sweep

glm53-flash-nvfp4-dgxspark-sglang-tp2-sweep

glm53-flash-nvfp4-dgxspark-sglang-tp2-sweep

Record

Recipe
glm53-flash-nvfp4-dgxspark-sglang-tp2
Measured
2026-08-27
Accepted
2026-08-27
Points
2

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatus
1131,07217.5accepted
2131,07215.4accepted

Remaining fields

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

accepted at
2026-08-27
id
glm53-flash-nvfp4-dgxspark-sglang-tp2-sweep
measured at
2026-08-27
recipe id
glm53-flash-nvfp4-dgxspark-sglang-tp2
schema version
local-ai-registry/v1

metrics

concurrency
2
latest point at
2026-08-27
max context tokens
131,072
peak generation tps
30.72
point count
2

rows

concurrencycontext tokensdecode tok sdecode tok s per streamoutput tokenspeak power wpeak temperature cpeak vram gb
1131,07217.4917.4913434.5153Unknown
2131,07230.7215.3654837.7456Unknown
Provenance & metadata (1)

source

commit
dfb1bb7e45c20058d37df7a39cceda45a9d216a8
paths
Dockerfile, compose.yaml, launch-spark.sh, manifest.json, chat-template-mm.jinja, patches/