# OpenMono.ai Inference Benchmark
# Test mode : cpu
# Generated : 2026-04-18T21:55:50Z
# Hostname  : echo1
# Endpoint  : http://localhost:7474
# Script    : cpu-test.sh
# Repo      : a5411f2

## Hardware

### CPU
Model: AMD Ryzen 9 7940HS w/ Radeon 780M Graphics
Physical cores: 8
Logical threads: 16

### Memory
Total: 26Gi   Used: 11Gi   Free: 297Mi

### OS
OS: Ubuntu 25.10
Kernel: 6.17.0-22-generic
Architecture: x86_64

### GPU
No NVIDIA GPU detected

## Model (from llama-server /props)

Model path: /models/qwen3.6-35b-a3b-ud-q4_k_xl.gguf
Context size: 32768

## Benchmark config

Iterations per test : 3
Warmup enabled      : yes
Temperature         : 0 (greedy)
top_k               : 1
Seed                : 42

## Per-iteration results

test            it    prefill_n   prefill/s    decode_n    decode/s     wall_ms
----            --    ---------   ---------    --------    --------     -------

### short-gen — Short decode (short prompt, 128 tokens out)

  short-gen         1          36  88.47295676621513         128  16.142060218712306        8358
  short-gen         2          36  88.34355828220859         128  17.166976745184126        7912
  short-gen         3          36  83.07238604660822         128  17.158487452520923        7915

### long-gen — Sustained decode (short prompt, 512 tokens out)

  long-gen          1          36  85.44493074213669         168  16.8685436402277       10402
  long-gen          2          36  86.48233117928268         168  16.698275773813506       10525
  long-gen          3          36  86.22775144371603         168  16.701833304569064       10511

### long-prefill — Prefill-heavy (long prompt, 128 tokens out)

  long-prefill      1         475  98.63964508002063         128  15.936064509189134       13036
  long-prefill      2         475  105.13983265722888         128  16.57367926227446       12262
  long-prefill      3         475  104.22513330943104         128  16.5874507900227       12296

### combined — Combined (long prompt, 512 tokens out)

  combined          1         475  103.63842886760071         512  15.831751523875656       36962
  combined          2         475  98.84492936957537         512  16.74668441815215       35454
  combined          3         475  100.52822822066517         512  16.354866879369375       36051

## Per-test summary (tokens/sec)

test            metric         min       max      mean    median
----            ------         ---       ---      ----    ------
short-gen       prefill      83.07     88.47     86.63     88.34
short-gen       decode       16.14     17.17     16.82     17.16
long-gen        prefill      85.44     86.48     86.05     86.23
long-gen        decode       16.70     16.87     16.76     16.70
long-prefill    prefill      98.64    105.14    102.67    104.23
long-prefill    decode       15.94     16.59     16.37     16.57
combined        prefill      98.84    103.64    101.00    100.53
combined        decode       15.83     16.75     16.31     16.35

## Raw data (TSV)

```tsv
test	iteration	prefill_tokens	prefill_per_sec	decode_tokens	decode_per_sec	wall_ms
short-gen	1	36	88.47295676621513	128	16.142060218712306	8358
short-gen	2	36	88.34355828220859	128	17.166976745184126	7912
short-gen	3	36	83.07238604660822	128	17.158487452520923	7915
long-gen	1	36	85.44493074213669	168	16.8685436402277	10402
long-gen	2	36	86.48233117928268	168	16.698275773813506	10525
long-gen	3	36	86.22775144371603	168	16.701833304569064	10511
long-prefill	1	475	98.63964508002063	128	15.936064509189134	13036
long-prefill	2	475	105.13983265722888	128	16.57367926227446	12262
long-prefill	3	475	104.22513330943104	128	16.5874507900227	12296
combined	1	475	103.63842886760071	512	15.831751523875656	36962
combined	2	475	98.84492936957537	512	16.74668441815215	35454
combined	3	475	100.52822822066517	512	16.354866879369375	36051
```