Browse runs

List runs

curl "$API_BASE/v1/training?page=1&per_page=25&sort_column=accuracy_score&sort_direction=desc" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

from predictai import PredictAI

client = PredictAI(token="pa_live_…", workspace_id="ws_…")

data = client.trainings.get_training(params={
    "page": 1,
    "per_page": 25,
    "sort_column": "accuracy_score",
    "sort_direction": "desc",
})

import { PredictAI } from "@predictai/sdk";

const client = new PredictAI({ token: "pa_live_…", workspaceId: "ws_…" });

const data = await client.trainings.getTraining({
  params: {
    page: 1,
    per_page: 25,
    sort_column: "accuracy_score",
    sort_direction: "desc",
  },
});

{
  "data": {
    "trainings": [
      {
        "uid": "7d81c3f0-…",
        "model": "",
        "blueprint_name": "",
        "created_at": "2026-07-14T02:00:04+00:00",
        "duration": 312.4,
        "cost": 40,
        "cost_unit": "credits",
        "executed": true,
        "message": "Training completed successfully",
        "accuracy_score": 0.94,
        "skill_score": 0.61,
        "comparison_metrics": "",
        "status": [
          { "action": "queued", "timestamp": "2026-07-14T02:00:04+00:00" },
          { "action": "started", "timestamp": "2026-07-14T02:00:19+00:00" },
          { "action": "completed", "timestamp": "2026-07-14T02:05:31+00:00" }
        ],
        "individual_comparisons": [
          { "ground_truth": [1201.5, 1187.0, …], "predicted": [1195.2, 1190.8, …] }
        ],
        "model_id": "9f2c41d8-…",
        "model_name": "Daily sales forecaster",
        "owner": "USER_ID",
        "last_status": "completed",
        "progress": null,
        "started_at": "2026-07-14T02:00:19+00:00",
        "completed_at": "2026-07-14T02:05:31+00:00",
        "adapter_id": "adp_9c2f…",
        "ready_for_promotion": true,
        "training_kind": "foundation",
        "report_summary": {
          "avg_train_loss": 0.0231,
          "rank": 8,
          "size_mb": 12.4,
          "family": "chronos_bolt",
          "backbone_slug": "base",
          "adapter_id": "adp_9c2f…"
        },
        "deployments": [
          { "promotion_id": "c4a7…", "deployment_type": "shared", "status": "deployed" }
        ]
      }
    ],
    "pagination": {
      "total_items": 38,
      "total_pages": 2,
      "current_page": 1,
      "per_page": 25
    },
    "total_count": 38,
    "page": 1,
    "per_page": 25,
    "total_pages": 2
  }
}

In training rows, model_id / model_name refer to the pipeline the run belongs to, and model is the catalog model's ID (blank for most foundation runs). A non-empty deployments array means the run is currently serving.

Query parameters

Parameter	Meaning
`page` optional	Page number (1-indexed). Defaults to `1`.
`per_page` optional	Rows per page. Defaults to `100`.
`sort_column` optional	Field to sort by; `accuracy_score` sorts on the report score. Defaults to `created_at`.
`sort_direction` optional	`asc` or `desc` (default).
`model_id` optional	Filter to a single pipeline.
`model_ids` optional	Comma-separated pipeline IDs.
`segment_ids` optional	Comma-separated segment IDs — matches runs from any pipeline training on those segments.
`statuses` optional	Comma-separated `last_status` values, e.g. `completed,failed`.

Filters combine: passing both model_ids and segment_ids returns only runs that satisfy both.

Inspect one run

curl "$API_BASE/v1/training/$TRAINING_ID" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

from predictai import PredictAI

client = PredictAI(token="pa_live_…", workspace_id="ws_…")

data = client.trainings.get_training_by_training_id(TRAINING_ID)

import { PredictAI } from "@predictai/sdk";

const client = new PredictAI({ token: "pa_live_…", workspaceId: "ws_…" });

const data = await client.trainings.getTrainingByTrainingId(TRAINING_ID);

{
  "data": {
    "uid": "7d81c3f0-…",
    "model_id": "9f2c41d8-…",
    "model_info": {
      "uid": "9f2c41d8-…",
      "name": "Daily sales forecaster",
      "description": "",
      "model": "",
      "segment": "SEGMENT_ID",
      "in_training": false,
      "forecast_horizon": 14
    },
    "model": "",
    "blueprint_info": {},
    "workspace_id": "WORKSPACE_ID",
    "owner": "USER_ID",
    "created_at": "2026-07-14T02:00:04+00:00",
    "started_at": "2026-07-14T02:00:19+00:00",
    "completed_at": "2026-07-14T02:05:31+00:00",
    "duration": 312.4,
    "cost": 40,
    "cost_unit": "credits",
    "framework": "PyTorch",
    "model_type": "Foundation LoRA (Chronos Bolt / base)",
    "model_size": "12.40 MB",
    "executed": true,
    "message": "Training completed successfully",
    "status": [ … ],
    "last_status": "completed",
    "progress": null,
    "queued": false,
    "training_kind": "foundation",
    "ready_for_promotion": true,
    "adapter_id": "adp_9c2f…",
    "adapter_info": {
      "kind": "foundation",
      "adapter_id": "adp_9c2f…",
      "family": "chronos_bolt",
      "backbone_slug": "base",
      "intensity": "adapt",
      "rank": 8,
      "lookback": 512,
      "horizon": 14,
      "size_mb": 12.4,
      "avg_train_loss": 0.0231,
      "n_batches": 400,
      "visibility": "private",
      "created_at": "2026-07-14T02:05:29+00:00"
    },
    "report": {
      "accuracy_score": 0.94,
      "skill_score": 0.61,
      "comparison_metrics": "",
      "individual_comparisons": [ … ],
      "feature_importance": [],
      "model_parameters": {},
      "training_metrics": {},
      "epochs": 0,
      "final_loss": 0,
      "metrics": { "avg_train_loss": 0.0231, "n_batches": 400 },
      "avg_train_loss": 0.0231,
      "rank": 8,
      "size_mb": 12.4,
      "family": "chronos_bolt",
      "backbone_slug": "base",
      "adapter_id": "adp_9c2f…",
      "val_mse": 1843.2,
      "val_mae": 31.7,
      "val_n_windows": 12,
      "val_quantile_used": 0.5
    },
    "configuration": {},
    "environment": {},
    "epochs_status": [],
    "has_epochs": false,
    "deployments": [
      { "promotion_id": "c4a7…", "deployment_type": "shared", "status": "deployed" }
    ]
  }
}

The detail response resolves everything a run touched — pipeline (model_info), catalog model (blueprint_info), fine-tuned adapter (adapter_info) — inline, so you never need a second call. Custom-model runs additionally populate epochs_status and has_epochs: true; see Concepts for how to read the scores.

Download the trained model

Every non-foundation run's artifact is yours to export — your models are never locked in:

curl "$API_BASE/v1/training/$TRAINING_ID/download" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

from predictai import PredictAI

client = PredictAI(token="pa_live_…", workspace_id="ws_…")

data = client.trainings.get_training_by_training_id_download(TRAINING_ID)

import { PredictAI } from "@predictai/sdk";

const client = new PredictAI({ token: "pa_live_…", workspaceId: "ws_…" });

const data = await client.trainings.getTrainingByTrainingIdDownload(TRAINING_ID);

{
  "data": {
    "download_url": "https://celmind-models.s3.amazonaws.com/…&X-Amz-Expires=3600…",
    "filename": "7d81c3f0-4a2e-4c9b-9f10-2b8d5e6a7c31.onnx",
    "format": "sklearn",
    "expires_in_seconds": 3600
  }
}

The URL is valid for one hour; request a fresh one any time. The file saves under the training's ID (the link carries a Content-Disposition header, so the name is applied automatically). What you get depends on how the model was trained:

Training	File
Custom — scikit-learn / XGBoost	the serialized `model.onnx` (served format)
Custom — PyTorch	the TorchScript `model.pt`
Custom — TensorFlow	the SavedModel directory, zipped (`….savedmodel.zip`, built on first request and cached)
BYOM	the original file you uploaded

Foundation runs are not downloadable: a fine-tuning run produces a small adapter on top of the foundation model's shared pretrained weights, which aren't a per-training artifact. Those requests return 403 not_downloadable.

Status	Why	Example message
`403`	Foundation training	`error.code` `not_downloadable`
`404`	Artifact gone from storage (cleaned up, or the run predates artifact storage)	`"Model artifact not found in storage…"`
`409`	Run still in progress or failed before serialization	`error.code` `not_downloadable`, includes `last_status`

Errors

The list and detail routes return errors as { "status": "<message>" }:

Status	Why	Example message
`400`	`X-Workspace-Id` header missing	`"Workspace ID is missing"`
`401`	Token missing or invalid	`"User context not found"`
`403`	No access to this workspace	`"Access denied to this workspace"`
`404`	Training doesn't exist in this workspace	`"Training with ID … not found"`

List runs

Query parameters

Inspect one run

Download the trained model

Errors

On this page