cellmap_flow.dashboard.routes.models

Attributes

`logger`
`models_bp`

Functions

`get_model_config_types`()	Get available ModelConfig subclasses and their parameter metadata
`create_model_config`()	Build the model config the form describes (a CreateModelConfig: the
`get_huggingface_models`()	Get available models from Hugging Face (uses cache if available)
`refresh_huggingface_models_route`()	Force refresh the Hugging Face models cache
`get_bioimage_models`()	The BioImage Model Zoo's models (cached): bioimage_catalog.list_bioimage_models.
`refresh_bioimage_models_route`()	Fetch the zoo's index again.
`submit_models`()	Run the models the Models tab has ticked (a SubmitModels), and stop
`get_loaded_models`()	The Models tab's Loaded models list: every model running or starting,
`remove_loaded_model`()	The ✕ of one model on the Loaded models list, `{"name": ...}`: stop
`job_logs`()	The inference jobs' own output, so a failure can be read here.
`gpu_queues`()	Which GPU queues are open and how busy, for the queue picker.
`get_server_config`()	Get current server configuration.
`export_config`()	Export the dashboard's current live config (models, normalization,
`update_server_config`()	Update server configuration and save to cache.

Module Contents

cellmap_flow.dashboard.routes.models.logger

cellmap_flow.dashboard.routes.models.models_bp

cellmap_flow.dashboard.routes.models.get_model_config_types(): Get available ModelConfig subclasses and their parameter metadata

cellmap_flow.dashboard.routes.models.create_model_config(): Build the model config the form describes (a CreateModelConfig: the type and its parameters, as a model entry) and keep it among the session’s models, for the pipeline builder and blockwise. A type or value the registry cannot build from is a 400.

cellmap_flow.dashboard.routes.models.get_huggingface_models(): Get available models from Hugging Face (uses cache if available)

cellmap_flow.dashboard.routes.models.refresh_huggingface_models_route(): Force refresh the Hugging Face models cache

cellmap_flow.dashboard.routes.models.get_bioimage_models(): The BioImage Model Zoo’s models (cached): bioimage_catalog.list_bioimage_models.

cellmap_flow.dashboard.routes.models.refresh_bioimage_models_route(): Fetch the zoo’s index again.

cellmap_flow.dashboard.routes.models.submit_models(): Run the models the Models tab has ticked (a SubmitModels), and stop the others: services.launch.update_run_models, each catalog and Hugging Face model at its run_voxel_sizes entry, if it has one. A zoo model that needs a voxel size and was given none, a Cellpose model without one, or a run voxel size for a model that reads its input itself, is a 400, and nothing changes. A chain with no input normalizers takes the ones the first model started declares, if any (get_loaded_models says so to the page).

cellmap_flow.dashboard.routes.models.get_loaded_models()

The Models tab’s Loaded models list: every model running or starting, with where it came from and whether its ✕ can remove it (services.launch.loaded_models), and the chain they run through.

{"success": true, "models": [...], "pipeline": {"input_norm", "postprocess"}, "input_norm_from", "input_norm_declared", "postprocess_from", "postprocess_fitted"}: pipeline is the chain as it is set (PUT /api/pipeline’s answer gives it the same way), input_norm_from the loaded model whose start set its input part as it is, else None, and input_norm_declared whether that model declares it (False: the convention for its kind of model; None with no such model; routes.pipeline.input_norm_set_by). postprocess_from is the loaded model whose settings its postprocessing is, else None, and postprocess_fitted whether the ranking fitted them to the paint (False: the model’s start set the sigmoid its logits need; None with no such model; routes.pipeline.postprocess_set_by). The page reads them after every model change: a model’s start can set the input normalization (routes.pipeline.set_declared_input_norm) and the postprocessing (set_needed_postprocess), and the Processing tab’s list then shows it, saying from which model and on what grounds. deleted_in_viewer lists the models stopped because their layer was deleted in the viewer, numbered in the order they went (the session’s record: services.launch._layers_deleted), for the list to untick the rows of those newer than it has seen, as their ✕ would.

cellmap_flow.dashboard.routes.models.remove_loaded_model(): The ✕ of one model on the Loaded models list, {"name": ...}: stop that model and take its layer away, and nothing else (services.launch.remove_model): a finetune job’s model’s job is cancelled with it. Answers the list as it is then; a 404 when no model of that name is running, a 409 when a finetune job could not be cancelled.

cellmap_flow.dashboard.routes.models.job_logs()

The inference jobs’ own output, so a failure can be read here.

Without this, a server that dies on startup or 500s on every chunk says nothing in the dashboard – the traceback is in the LSF job’s output on a cluster node, and reading it means logging in and running bpeek.

cellmap_flow.dashboard.routes.models.gpu_queues()

Which GPU queues are open and how busy, for the queue picker.

Polled about once a minute by the models tab; the underlying LSF query is cached server-side, so this is cheap to call.

cellmap_flow.dashboard.routes.models.get_server_config(): Get current server configuration.

cellmap_flow.dashboard.routes.models.export_config(): Export the dashboard’s current live config (models, normalization, postprocessing, queue/charge_group) as a downloadable YAML file that can be reloaded later with cellmap_flow yaml.

cellmap_flow.dashboard.routes.models.update_server_config(): Update server configuration and save to cache.