The json format is an analogy, and we certainly do not have to strictly use a
json file or string. For example, in the python API, we could directly from a
dictionary(in the style of json).
```python
target = tvm.target.create({
"id": "cuda",
"target_host": {"id" : "llvm"}
});
```
We can also put first level API wrappers around to create the object.
```python
target = tvm.target.cuda(target_host="xyz")
```
Alternatively, we sgiykd allow user to directly use the tag to create a target.
e.g.
```python
target = tvm.target.create("nvidia/gtx1080ti")
```
The schema registration and validation are exactly designed to tame the
complexity. While right now we only focuses on validation, we could certainly
imagine generating docs that include these options. Alternatives, we can simply
document the corresponding fields in a specific doc about the targets or as
arguments of the convenient functions(e.g. tvm.target.cuda), love to hear more
thoughts.
I am not particularly attached to the choice of id, so feel free to suggest
alternatives.
---
[Visit Topic](https://discuss.tvm.ai/t/rfc-tvm-target-specification/6844/10) to
respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.ai/email/unsubscribe/5b485d09236f75c5fb08b2401ad1ce884d87848a4b240be92135681bb706cced).
