Function remote_model

Connect to a remote model as if it were a local differentiable function.

Remarks

This function assumes a server running network.serve_model.

Example

Similar to network.tfetch, the await keyword may be used:

const model = sm.network.remote_model('localhost:3000')
const input = sm.randn([128])
// await is necessary, as model invocation is asynchronous
const output = await model(input)
const loss = sm.loss.mse(output, reference)
// note the use of the await keyword for the backward pass
await loss.backward()

or chained callbacks:

model(input).then((output) => {
  loss_fn(output).backward()
}).then(() => {
  console.log('backward executed!')
})

Returns

An asynchronous function that can be invoked with an input tensor.