add execute method for inference model to allow retrieve activation for any internal nodes

[pyu10055]

This change is 

[caisq]

Review status: 0 of 1 files reviewed at latest revision, all discussions resolved.

---

src/types.ts, line 167 at r1 (raw file):

NamedTensorMap

Can we remove this option? The outputs will always be returned as a single Tensor or an Array of them. Otherwise it's unclear when the return value will be an array and when it'll be a dict (object).

---

src/types.ts, line 193 at r1 (raw file):

config: ModelPredictConfig,

Can we remove this? See my comment in the doc. I think batching is at a level higher than execute().

---

Comments from Reviewable

[pyu10055]

Review status: 0 of 1 files reviewed at latest revision, 2 unresolved discussions.

---

src/types.ts, line 167 at r1 (raw file):

Previously, caisq (Shanqing Cai) wrote…

NamedTensorMap

Can we remove this option? The outputs will always be returned as a single Tensor or an Array of them. Otherwise it's unclear when the return value will be an array and when it'll be a dict (object).

it is needed for frozen model, since the order of output is not explicit.

---

src/types.ts, line 193 at r1 (raw file):

Previously, caisq (Shanqing Cai) wrote…

config: ModelPredictConfig,

Can we remove this? See my comment in the doc. I think batching is at a level higher than execute().

Done.

---

Comments from Reviewable

[caisq]

Thanks!

---

Review status: 0 of 1 files reviewed at latest revision, 2 unresolved discussions.

---

src/types.ts, line 178 at r2 (raw file):

 Optional. 

This is not optional, right? Also, there is no default, right?

---

src/types.ts, line 183 at r2 (raw file):

multiple outputs.

Add that the order is the same as outputs.

---

Comments from Reviewable