File size: 4,197 Bytes
336c50f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
{
  "passed": true,
  "model": "mirrorneuronlab/embeddinggemma-2:q4_k_m",
  "endpoint": "http://127.0.0.1:12435/engines/llama.cpp/v1/embeddings",
  "text": {
    "http_status": 200,
    "elapsed_seconds": 1.8947369380039163,
    "vectors": 4,
    "dimensions": 768,
    "finite": true,
    "l2_norms": [
      0.9999999906183037,
      0.9999999717982236,
      0.9999999449210023,
      0.9999999732160483
    ],
    "related_cosine": 0.8650823975051903,
    "unrelated_cosine": 0.5486270130835971
  },
  "repeat": {
    "http_status": 200,
    "elapsed_seconds": 0.03770295198773965,
    "vectors": 4,
    "dimensions": 768,
    "finite": true,
    "l2_norms": [
      0.9999999906183037,
      0.9999999717982236,
      0.9999999449210023,
      0.9999999732160483
    ],
    "max_abs_difference": 0.0
  },
  "cosine_to_bf16_text": [
    0.9941200584690152,
    0.9908234124220315,
    0.9889631445265419,
    0.9972267675976894
  ],
  "image": {
    "http_status": 200,
    "elapsed_seconds": 0.12217680300818756,
    "vectors": 1,
    "dimensions": 768,
    "finite": true,
    "l2_norms": [
      1.0000000458974723
    ]
  },
  "audio": {
    "http_status": 200,
    "elapsed_seconds": 0.04252175899455324,
    "vectors": 1,
    "dimensions": 768,
    "finite": true,
    "l2_norms": [
      1.0000000058851288
    ]
  },
  "mixed_batch": {
    "http_status": 200,
    "elapsed_seconds": 0.06744719200651161,
    "vectors": 3,
    "dimensions": 768,
    "finite": true,
    "l2_norms": [
      0.9999999906183037,
      1.0000000458974723,
      1.0000000058851288
    ]
  },
  "scope": "Functional smoke checks on four texts, one solid-color image, and one synthetic audio tone. Not a retrieval or multimodal quality benchmark.",
  "docker": {
    "engine": "29.6.2",
    "client": "v1.2.9",
    "server": "v1.2.9",
    "image": "docker/model-runner@sha256:9967b6c1092ff01e7cf68119ab25067a1f58e3d1b06873bc16fe1e758bcd914f",
    "llama_cpp_revision": "dde879bc4f954b809892fd0d660515f42536394a",
    "llama_cpp_build": "b12032 / 0.8.0-dev",
    "cuda_architecture": "121",
    "local_model": "mirrorneuronlab/embeddinggemma-2:q4_k_m",
    "endpoint": "http://127.0.0.1:12435",
    "configuration": "-ngl 999 --pooling mean -b 2048 -ub 2048 --parallel 1 -t 10; context 2048; embedding mode",
    "gpu_evidence": "Docker logged gpuSupport=true, backend arguments include -ngl 999, NVIDIA-SMI reports /opt/llama/llama-server GPU processes.",
    "backend_version_note": "The stock images checked bundled older engines without this architecture. Validation used the official current v1.2.9 controller with the pinned upstream CUDA backend mounted read-only. The controller displays the backend version as unknown because its parser does not recognize llama.cpp's new version string."
  },
  "bf16_reference": {
    "http_status": 200,
    "elapsed_seconds": 1.0279945080110338,
    "vectors": 4,
    "dimensions": 768,
    "finite": true,
    "l2_norms": [
      1.0000000336112962,
      1.0000000038732915,
      0.9999999638015133,
      1.0000000373879046
    ],
    "related_cosine": 0.8629100120127821,
    "unrelated_cosine": 0.5291825480995416
  },
  "docker_model_run": {
    "stage": "dmr-run",
    "command": [
      "env",
      "MODEL_RUNNER_HOST=http://127.0.0.1:12435",
      "docker",
      "model",
      "run",
      "--detach",
      "mirrorneuronlab/embeddinggemma-2:q4_k_m"
    ],
    "cwd": "/home/homer/Sandbox/hf_model/embeddinggemma2-q4",
    "started_utc": "2026-11-01T15:05:52.619080+00:00",
    "finished_utc": "2026-11-01T15:05:52.670364+00:00",
    "elapsed_seconds": 0.0512839280127082,
    "returncode": 0
  },
  "fixtures": [
    {
      "name": "red-square.png",
      "sha256": "66d2ca44529ac40d3d08a58d0fd0f10696eae138d288dc5cfc1351f263d4c0a3",
      "size_bytes": 677
    },
    {
      "name": "tone-440hz.wav",
      "sha256": "3f806962fcdcc982acaca0e5180c701391a59c2635d82e4b0660dc148c01f528",
      "size_bytes": 64044
    }
  ],
  "other_platforms": "Apple Silicon Macs and Ubuntu/NVIDIA systems were not physically tested. GGUF files are platform portable; their llama.cpp backend must support gemma-embedding2 and the Gemma4 vision/audio encoders."
}