Speed sweep

glm-5-3-exl3-tr3-3-0bpw-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep

glm-5-3-exl3-tr3-3-0bpw-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep

Record

Recipe
glm-5-3-exl3-tr3-3-0bpw-rtx-pro-6000-blackwell-96gb-vllm-tp4
Measured
2026-09-01T03:45:33.045685Z
Accepted
2026-09-01T04:00:19Z
Points
3

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatus
1781,752.261.2325.5validated
2781,768.142.9503.1validated
3781,855.538.5675validated

Remaining fields

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

accepted at
2026-09-01T04:00:19Z
id
glm-5-3-exl3-tr3-3-0bpw-rtx-pro-6000-blackwell-96gb-vllm-tp4-sweep
measured at
2026-09-01T03:45:33.045685Z
recipe id
glm-5-3-exl3-tr3-3-0bpw-rtx-pro-6000-blackwell-96gb-vllm-tp4
schema version
local-ai-registry/v1

metrics

concurrency
3
decode mode
sustained-matched-window
inference engine version
0.26.1rc0+infernal.invocation.cu133.r17.vllmc53cc73.b12xc0a44a1
latest point at
2026-09-01T03:45:33.045685Z
max context tokens
800,000
max prompt tokens
8,192
peak generation tps
115.542
peak memory bytes
397263503360
peak prompt tps
1,855.53
point count
3

rows

aggregate sourcecapacity limitedconcurrencycontext tokensdecode tok sdecode tok s per streameffective concurrencyerrors
openai_continuous_usageNo17861.21161.21110
openai_continuous_usageNo27885.78342.89120
openai_continuous_usageNo378115.54238.51430
Provenance & metadata (3)

facts

metrics.peak generation tps · provenance · captured at
2026-09-01T04:00:19Z
metrics.peak generation tps · reason openai-continuous-usage-completion-tokens-divided-by-common-measurement-window
metrics.peak generation tps · state known
metrics.peak memory bytes · provenance · captured at
2026-09-01T04:00:19Z
metrics.peak memory bytes · reason maximum-summed-memory-used-from-340-half-second-four-gpu-samples
metrics.peak memory bytes · state known
metrics.peak prompt tps · provenance · captured at
2026-09-01T04:00:19Z
metrics.peak prompt tps · reason exact-prompt-tokens-divided-by-common-wall-time-to-first-token-with-uncapped-natural-stop-requests
metrics.peak prompt tps · state known

provenance

captured at
2026-09-01T04:00:19Z

source

kind
live-gpu-benchmark
paths
docs/notes/glm-5-3-exl3-tr3-3-0bpw-rtx-pro-6000-blackwell-20260901.json