onnx-genai-example-gemma4-e2b / runtime_versions.json
justinchuby's picture
Publish metadata-complete Gemma-4 E2B-it text decoder (fp16 CUDA): canonical inference_metadata.yaml + policies + L4/L5 parity; supersedes prior metadata-less multimodal export
23835eb verified
Raw
History Blame Contribute Delete
545 Bytes
{
"onnxruntime": "1.27.0 (onnxruntime-gpu)",
"execution_provider": "CUDAExecutionProvider (NVIDIA H200)",
"onnx_ir": "1.0.0",
"onnxscript": "0.7.1",
"transformers": "5.14.1",
"torch": "2.13.0",
"python": "3.12",
"dtype": "float16",
"mobius_git_sha": "0776f562daeefad468277177b1e92abb26486355",
"build_flags": "--features text-only --dtype f16 --ep cuda --optimize=group_query_attention,packed_attention,skip_norm",
"package_kind": "target_decoder",
"canonical_metadata": "inference_metadata.yaml (schema 1.0, hashless)"
}