Speed sweep

deepseek-v4-flash-0731-ud-iq2-m-rtx-3090-24gb-llama-cpp-tp4-sweep

deepseek-v4-flash-0731-ud-iq2-m-rtx-3090-24gb-llama-cpp-tp4-sweep

Record

Recipe
deepseek-v4-flash-0731-ud-iq2-m-rtx-3090-24gb-llama-cpp-tp4
Measured
2026-08-06T21:14:50.076Z
Points
1

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatus
1131,0721,745.441.118,774observed

Remaining fields

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

id
deepseek-v4-flash-0731-ud-iq2-m-rtx-3090-24gb-llama-cpp-tp4-sweep
measured at
2026-08-06T21:14:50.076Z
recipe id
deepseek-v4-flash-0731-ud-iq2-m-rtx-3090-24gb-llama-cpp-tp4
schema version
local-ai-registry/v1

metrics

concurrency
1
inference engine version
69 (fork ds4-longctx @ e6718011c)
latest point at
2026-08-06T21:14:50.076Z
max context tokens
131,072
peak generation tps
41.07
peak prompt tps
1,745.4
point count
1

rows

concurrencycontext tokensdecode tok sdecode tok s per streamoutput tokenspeak vram gbprefill tok ssamples
1131,07241.0741.0725690.31,745.41
Provenance & metadata (1)