cellmap_flow.serving.protocol

What passes between the dashboard and an inference server, as text.

A prediction layer’s source is zarr://<server>/<model><ARGS_KEY><blob><ARGS_KEY>. The blob is the layer’s chain, {INPUT_NORM_KEY: [...], POSTPROCESS_KEY: [...]} followed by extras such as dashboard_url and digest, as minified JSON in URL-safe base64 without padding (encode_to_str). Every request the viewer makes carries it, and split_dataset_url finds it in the dataset part of a request’s path. pipeline_spec.PipelineSpec reads and writes the chain; this module is only the codec, so a server and a dashboard of different versions still understand each other.

The other way, a server announces its address by printing it between the two IP_PATTERN markers (CellMapFlowServer.run). They are defined in jobs.spec, which finds them in a job’s log.

The same codec carries a dict argument on a command line: a model entry for the finetune CLI’s --model-entry, and a finetuned model’s base_model for cellmap_flow_server.

Attributes

`ARGS_KEY`
`INPUT_NORM_KEY`
`POSTPROCESS_KEY`

Functions

`encode_to_str`(data)	Encodes a JSON object into a URL-safe string without '/', '+', or '='.
`decode_to_json`(encoded_str)	Decodes a URL-safe string back into a JSON object.
`split_dataset_url`(→ Optional[str])	The args blob between a layer URL's two `ARGS_KEY` markers.

Module Contents

cellmap_flow.serving.protocol.ARGS_KEY = '__CFLOW_ARGS__'

cellmap_flow.serving.protocol.INPUT_NORM_KEY = 'input_norm'

cellmap_flow.serving.protocol.POSTPROCESS_KEY = 'postprocess'

cellmap_flow.serving.protocol.encode_to_str(data): Encodes a JSON object into a URL-safe string without ‘/’, ‘+’, or ‘=’.

cellmap_flow.serving.protocol.decode_to_json(encoded_str): Decodes a URL-safe string back into a JSON object.

cellmap_flow.serving.protocol.split_dataset_url(dataset: str) → str | None

The args blob between a layer URL’s two ARGS_KEY markers.

None when the URL carries no blob at all. Any other number of markers than two is a malformed URL.