WACS.WASI.NN.OnnxRuntimeGenAI
OnnxRuntime-GenAI backend for WACS.WASI.NN. Wraps Microsoft.ML.OnnxRuntimeGenAI for first-class generative LLM inference (Gemma 3, Llama, Qwen, Phi families) through wasi-nn. Exposes two compute shapes selected by named-input convention: "prompt" (utf-8) runs full tokenize -> KV-cached decode -> detokenize in one call; "input_ids" (int64) runs a single forward pass returning logits for guests that drive their own decode loop. Linked against CoreML.framework on osx-arm64 for Metal-capable acceleration where the underlying ORT EP supports it.
View on Github
Activity
- Latest release
- 4mo ago
- Total releases
- 3
- Cadence
- ~daily
- Last 12 months
- 3
Reach
- Stars
- —
Details
- License
- Apache-2.0
- First release
- May 11, 2026
Releases
| Version | Released | |
|---|---|---|
0.1.5
patch
| ||
0.1.4
patch
| ||
0.1.3
initial
|