embeddinggemma-2-GGUF / VALIDATION.json
homerquan's picture
Add standard Q4_K_M text model and Q8_0 vision/audio companion
336c50f verified
Raw History Blame Contribute Delete
4.2 kB
{
"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."
}