--- license: apache-2.0 tags: - onnx - onnxruntime - onnx-genai - inference-metadata --- # onnx-genai-example-qwen2-5-0-5b-cuda-gqa-f16 Private real-weight ONNX package produced by Mobius from [`Qwen/Qwen2.5-0.5B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct/tree/7ae557604adf67be50417f59c2c2f167def9a775) at immutable revision `7ae557604adf67be50417f59c2c2f167def9a775`. Source license: `apache-2.0`. This package contains 24 CUDA `com.microsoft::GroupQueryAttention` nodes. `weight_integrity.json` validates every packed QKV initializer against the pinned source weights. ## Contents - Canonical, hashless `inference_metadata.yaml` - ONNX graphs and external-data weights - Complete tokenizer/processor assets - `request.json` and `output.json` real runtime evidence - `graph_report.json`, `performance.json`, `source.json`, and `provenance.json` Observed output: `Write one short sentence about ONNX. ONNX is a popular open-source framework` ## Exact download ```bash hf download justinchuby/onnx-genai-example-qwen2-5-0-5b-cuda-gqa-f16 --repo-type model --local-dir ./qwen2.5-0.5b-cuda-gqa-f16 ``` ## ONNX Runtime load smoke test ```bash python - <<'PY' from pathlib import Path import onnxruntime as ort root = Path("qwen2.5-0.5b-cuda-gqa-f16") for relative_path in ['model.onnx']: session = ort.InferenceSession( str(root / relative_path), providers=['CUDAExecutionProvider', 'CPUExecutionProvider'], ) print(relative_path, session.get_providers(), [x.name for x in session.get_inputs()]) PY ``` The exact successful probe request, output, versions, providers, and timings are preserved in `request.json`, `output.json`, and `performance.json`. ## Annotated inference metadata Review [`inference_metadata.annotated.yaml`](./inference_metadata.annotated.yaml) for inline explanations of this package's workflow, tensor/state/cache contracts, and fail-closed omissions. [`inference_metadata.yaml`](./inference_metadata.yaml) remains the canonical machine-authored contract; automated validation confirms both files parse to the same metadata object.