可复现实测

先比较实验条件，再比较结果

这里保留公开结果的上下文和缺失项，帮助团队建立自己的容量基线。

不可直接横向比较

RTFx 会随硬件、音频分布、VAD、并发、预热和计时边界变化。缺少任一条件的结果不能作为采购或容量承诺。

固定输入

记录音频总时长、文件分布、语言、声道和采样率。

记录 CPU/GPU、驱动、模型、提交、运行时和所有参数。

说明是否包含加载、预热、I/O、排队和后处理。

公开记录

vLLM GPU 高吞吐

RTFx 340; CER 8.20%

运行时: vLLM batch
硬件: GPU model not recorded in the cited public table
负载: Offline batch
音频: 184 long-form files; 11,541 seconds
设置: Dynamic VAD; the public table does not record batch size or the full software and hardware stack
计时口径: Offline throughput; the cited table does not state whether warmup, file I/O, and decoding are excluded
核对日期: 2026-07-26

限定：Incomplete hardware and timing record. Use only as public reference evidence, not a capacity promise.

llama.cpp / GGUF 独立运行

Approximately 20x realtime; Q8 CER 8.17%

运行时: llama.cpp/GGUF with built-in FSMN-VAD
硬件: CPU, 8 threads; CPU model not recorded in the cited report
负载: Offline Mandarin CPU
音频: 184 Mandarin clips, about 44-60 seconds each; total duration not recorded in the cited report
设置: Q8 runtime; built-in FSMN-VAD; micro-CER normalize_zh; model load excluded
计时口径: Sum of compute time divided by sum of audio duration; model load excluded
核对日期: 2026-07-26

限定：CPU model and full software stack are not recorded. Reproduce on target hardware before comparison.

llama.cpp / GGUF 独立运行

Approximately 21x realtime; Q8 CER 9.89%

运行时: llama.cpp/GGUF with built-in FSMN-VAD
硬件: CPU, 8 threads; CPU model not recorded in the cited report
负载: Offline Mandarin CPU
音频: 184 Mandarin clips, about 44-60 seconds each; total duration not recorded in the cited report
设置: Q8 runtime; built-in FSMN-VAD; micro-CER normalize_zh; model load excluded
计时口径: Sum of compute time divided by sum of audio duration; model load excluded
核对日期: 2026-07-26

限定：CPU model and full software stack are not recorded. Reproduce on target hardware before comparison.

实时服务

文件吞吐不能说明首稿延迟、最终结果延迟、长连接并发和背压。使用实时 WebSocket 方法记录这些指标。