cellmap_flow.dashboard.finetune_layers

What the dashboard shows of a finetune job’s model as it trains.

The job manager (finetune.job_manager) knows nothing of the viewer: it tells its listeners when a job’s inference server is up and when an iteration finishes (FinetuneJobListener). The dashboard’s listener, FinetuneLayerListener, answers each event with the dashboard’s two views of the model:

• add_finetuned_layer(): its viewer layer, served by the job’s server, replacing the previous iteration’s; and its job among the session’s jobs;

• register_finetuned_model(): its FinetuneModelConfig among the session’s models, so the pipeline builder offers it.

follow_jobs(manager) has the listener told of a manager’s jobs; the routes that put jobs into the manager (submit, and finding the jobs of an earlier dashboard) call it first.

Attributes

logger

Classes

FinetuneLayerListener

The dashboard's listener: on each event, the model's layer and its

Functions

add_finetuned_layer(job, model_name)	Add (or replace) the layer of job's model, model_name.
register_finetuned_model(job, model_name)	Put a FinetuneModelConfig for job's model model_name among the
follow_jobs(manager)	Have the dashboard's listener told of manager's jobs, once however

Module Contents

cellmap_flow.dashboard.finetune_layers.logger

cellmap_flow.dashboard.finetune_layers.add_finetuned_layer(job, model_name)

Add (or replace) the layer of job’s model, model_name.

The layer is the one every model gets (viewer.layers.prediction_layer), over the job’s inference server, in red; the previous iteration’s layer (job.finetuned_model_name) goes. The session’s jobs get a job for the server in place of any earlier iteration’s, so Submit and the pipeline builder see the model as running. Nothing is added before the server is up.

cellmap_flow.dashboard.finetune_layers.register_finetuned_model(job, model_name)

Put a FinetuneModelConfig for job’s model model_name among the session’s models, in place of any earlier iteration’s, so the pipeline builder offers it with its parameters filled in.

class cellmap_flow.dashboard.finetune_layers.FinetuneLayerListener

The dashboard’s listener: on each event, the model’s layer and its pipeline-builder model. Either failing is logged, and does not stop the other.

on_server_ready(job, url, model_name)

The job’s inference server is up at url, serving model_name.

on_iteration_complete(job, model_name)

The job finished a training iteration and named its model model_name.

cellmap_flow.dashboard.finetune_layers.follow_jobs(manager)

Have the dashboard’s listener told of manager’s jobs, once however often this is called (request threads call it concurrently).