es-ude / es-ude/OnDeviceTraining
Implement InferenceImplementationProviders (Float32, SymInt32, Asym)
- Dominant language
- C
- Stars
- 1
- Forks
- 3
- Avg merge
- 1d 1h
- Merged PRs (30d)
- 8
Description
## Context
Sub-issue of #58.
**Blocked by:** #59 (scaffolding)
**Blocks:** #62 (ir2c), #64 (E2E tests)
## Goal
Implement `InferenceImplementationProvider` protocol for all three quantization types.
## Protocol (defined by creator)
```python
class InferenceImplementationProvider(Protocol):
def model_attributes(self, sub_graph: DataGraph) -> AttributeMapping: ...
```
Simpler than training — only `model_attributes()` with forward-pass C symbols.
## Acceptance Criteria
- [ ] `Float32InferenceProvider` implemented and tested
- [ ] `SymInt32InferenceProvider` implemented and tested
- [ ] `AsymInferenceProvider` implemented and tested
- [ ] Providers return valid attribute mappings accepted by `apply_inference_provider`
- [ ] C symbol names match actual function names in the C framework
Contributor guide
Assessment
This issue has not been assessed yet.