Speed sweep

qwen3-8-27b-uncensored-int4-autoround-gptq-int4-intel-arc-pro-b70-32gb-vllm-tp1-sweep

qwen3-8-27b-uncensored-int4-autoround-gptq-int4-intel-arc-pro-b70-32gb-vllm-tp1-sweep

Record

Recipe: qwen3-8-27b-uncensored-int4-autoround-gptq-int4-intel-arc-pro-b70-32gb-vllm-tp1
Measured: 2026-08-20T02:07:49.136Z
Points: 1

Measured speed

Concurrency	Context	Prefill	Decode	TTFT ms	Status
—	2,048	1,782.1	97.8	356.3	observed

Remaining fields

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

id: qwen3-8-27b-uncensored-int4-autoround-gptq-int4-intel-arc-pro-b70-32gb-vllm-tp1-sweep
measured at: 2026-08-20T02:07:49.136Z
recipe id: qwen3-8-27b-uncensored-int4-autoround-gptq-int4-intel-arc-pro-b70-32gb-vllm-tp1
schema version: local-ai-registry/v1

metrics

inference engine version: 0.27.2rc1.dev78+g8e6d8e4f6 XPU (custom-patched: host-staged TP2 collectives + INT8 W8A8 lm_head + MTP wiring)
latest point at: 2026-08-20T02:07:49.136Z
max context tokens: 2,048
peak generation tps: 97.8
peak prompt tps: 1,782.1
point count: 1

rows

concurrency	context tokens	decode tok s	decode tok s per stream	output tokens	peak vram gb	prefill tok s	samples
Unknown	2,048	97.8	Unknown	37	Unknown	1,782.1	1

Provenance & metadata (1)

source