Local AI Registry

Speed sweep

glm53-flash-nvfp4-rtxpro6000-sglang-tp4-sweep

glm53-flash-nvfp4-rtxpro6000-sglang-tp4-sweep
JSON APIBack to speed sweep search

Record

Recipe
glm53-flash-nvfp4-rtxpro6000-sglang-tp4
Measured
2026-08-26
Points
10

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatus
1——141.8124observed
2——123252observed
4——100.3256observed
8——72.5332observed
12,1255,344—398observed
16,3256,383—991observed
121,4076,275—3,411observed
153,3156,180—8,627observed
1106,7266,029—17,702observed
1213,0945,783—36,850observed

Remaining fields

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

id
glm53-flash-nvfp4-rtxpro6000-sglang-tp4-sweep
measured at
2026-08-26
recipe id
glm53-flash-nvfp4-rtxpro6000-sglang-tp4
schema version
local-ai-registry/v1

metrics

concurrency
8
latest point at
2026-08-26
max context tokens
213,094
peak generation tps
550
peak prompt tps
6,383
point count
10

rows

concurrencycontext tokensdecode tok sdecode tok s per streamoutput tokenspeak vram gbprefill tok ssamples
1Unknown142141.8512UnknownUnknown1
2Unknown240.41231,024UnknownUnknown2
4Unknown381.8100.32,048UnknownUnknown4
8Unknown55072.54,096UnknownUnknown8
12,125UnknownUnknown0Unknown5,3441
16,325UnknownUnknown0Unknown6,3831
121,407UnknownUnknown0Unknown6,2751
153,315UnknownUnknown0Unknown6,1801
1106,726UnknownUnknown0Unknown6,0291
1213,094UnknownUnknown0Unknown5,7831
Provenance & metadata (1)

source

commit
46ea34371842cfce939f4f76b5e60c1e99235f57
paths
Dockerfile, manifest.json, compose.yaml, README.md, patches/