DashboardClient

HTTP client for the EvolvingSystems backend's SDK write routes, with a local mirror and graceful degradation when the backend is unreachable.

Attributes

attributebase_url

= (base_url or os.environ.get(EVSYS_API_URL_ENV) or DEFAULT_API_URL).rstrip('/')

attributeapi_key

= api_key or os.environ.get(EVSYS_API_KEY_ENV)

attributeproject_id

= project_id or os.environ.get(EVSYS_PROJECT_ID_ENV)

attributetimeout_s

= timeout_s

attributeoffline

= offline if offline is not None else truthy_env(os.environ.get(EVSYS_OFFLINE_ENV))

attributelocal

= LocalExperimentStore(log_dir=log_dir)

Functions

func__init__(self, *, base_url=None, api_key=None, project_id=None, timeout_s=DEFAULT_TIMEOUT_S, offline=None, log_dir=None) -> None

paramself

parambase_urlstr | None

= None

paramapi_keystr | None

= None

paramproject_idstr | None

= None

paramtimeout_sfloat

= DEFAULT_TIMEOUT_S

paramofflinebool | None

= None

paramlog_dirstr | None

= None

Returns

None

func_post(self, path, body=None) -> dict | None

POST to the backend. Returns the JSON dict, or None when the write could not be sent remotely (offline, or backend unreachable). Raises DashboardClientError on a 4xx response.

paramself

parampathstr

parambodydict | None

= None

Returns

dict | None

funccreate_experiment

(self, *, experiment_name, hypothesis=None, hypothesis_reasoning=None, plan=None, tags=None, project_goal_id=None, config=None, declaration_path=None, is_valid=None) -> dict

paramself

paramexperiment_namestr

paramhypothesisstr | None

= None

paramhypothesis_reasoningstr | None

= None

paramplanstr | None

= None

paramtagslist[str] | None

= None

paramproject_goal_idstr | None

= None

paramconfigdict | None

= None

paramdeclaration_pathstr | None

= None

paramis_validbool | None

= None

Returns

dict

funcupdate_experiment(self, experiment_id, **patch) -> dict

PATCH experiment metadata. Whitelisted backend fields: status, best_score, best_generation_id, current_iteration, error_message, hypothesis, hypothesis_reasoning, plan, conclusion, tags, problem_statement_id, is_valid.

e.g. update_experiment(exp_id, is_valid=False) to invalidate an experiment after a bug is found in its runs (D16).

paramself

paramexperiment_idstr

parampatchAny

= {}

Returns

dict

funccreate_run

(self, *, experiment_id, group_id=None, dataset_id=None, seed=None, recipe_kind=None, run_config=None, status=STATUS_PENDING, wandb_run_url=None, tensorboard_path=None, tensorboard_archive_url=None) -> dict

paramself

paramexperiment_idstr

paramgroup_idstr | None

= None

paramdataset_idstr | None

= None

paramseedint | None

= None

paramrecipe_kindstr | None

= None

paramrun_configdict | None

= None

paramstatusstr

= STATUS_PENDING

paramwandb_run_urlstr | None

= None

paramtensorboard_pathstr | None

= None

paramtensorboard_archive_urlstr | None

= None

Returns

dict

funcupdate_run(self, run_id, **patch) -> dict

paramself

paramrun_idstr

parampatchAny

= {}

Returns

dict

funclog_step_metric

(self, run_id, *, step, split='train', loss=None, accuracy=None, learning_rate=None, grad_norm=None, tokens_per_sec=None, **metrics) -> dict

Log per-step training metrics as long-format run_metrics (D15).

Accepts arbitrary named series via **metrics (e.g. val_loss=…, kl_loss=…) plus a split (train/val/test). The named kwargs (loss, accuracy, …) are folded into the same metrics map. None values are dropped.

paramself

paramrun_idstr

paramstepint

paramsplitstr

= 'train'

paramlossfloat | None

= None

paramaccuracyfloat | None

= None

paramlearning_ratefloat | None

= None

paramgrad_normfloat | None

= None

paramtokens_per_secfloat | None

= None

parammetricsfloat

= {}

Returns

dict

funccreate_eval

(self, run_id, *, metrics, benchmark_id=None, checkpoint_id=None, model_ref=None, step=None, breakdowns=None, sdk_version=None) -> dict

Record an eval (D13): a run scored on a benchmark → metrics{name:value}.

paramself

paramrun_idstr

parammetricsdict[str, float]

parambenchmark_idstr | None

= None

paramcheckpoint_idstr | None

= None

parammodel_refstr | None

= None

paramstepint | None

= None

parambreakdownsdict | None

= None

paramsdk_versionstr | None

= None

Returns

dict

funcadd_checkpoint(self, run_id, *, uri, label=None, step=None, base_model=None, is_final=False) -> dict

Record a named checkpoint for a run (D7).

paramself

paramrun_idstr

paramuristr

paramlabelstr | None

= None

paramstepint | None

= None

parambase_modelstr | None

= None

paramis_finalbool

= False

Returns

dict

funclog_predictions(self, run_id, predictions) -> dict

Bulk-insert per-task predictions. Each is: { kind: 'eval' | 'rollout', eval_id?, task_id?, sample_idx?, step?, instruction, model_output, expected?, reward?, advantage?, metadata? }

paramself

paramrun_idstr

parampredictionslist[dict]

Returns

dict

DashboardClient

Attributes

Functions

On this page