scorebook 0.0.11__tar.gz → 0.0.13__tar.gz

{scorebook-0.0.11 → scorebook-0.0.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scorebook
-Version: 0.0.11
+Version: 0.0.13
 Summary: A Python project for LLM evaluation.
 License-File: LICENSE
 Author: Euan Campbell
@@ -37,7 +37,7 @@ Requires-Dist: torch ; extra == "examples"
 Requires-Dist: torchaudio ; extra == "examples"
 Requires-Dist: torchvision ; extra == "examples"
 Requires-Dist: transformers ; extra == "examples"
-Requires-Dist: trismik (>=1.0.1,<2.0.0)
+Requires-Dist: trismik (==1.0.2)
 Description-Content-Type: text/markdown
 
 # Scorebook

{scorebook-0.0.11 → scorebook-0.0.13}/pyproject.toml

@@ -11,7 +11,7 @@ requires-python = ">=3.9, <3.14"
 dependencies = [
     "datasets>=3.6.0",
     "notebook (>=7.4.5,<8.0.0)",
-    "trismik (>=1.0.1, <2.0.0)",
+    "trismik==1.0.2",
     "ipywidgets>=8.0.0",
 ]
 
@@ -19,14 +19,14 @@ dependencies = [
 scorebook = "scorebook.cli.main:main"
 
 [tool.poetry]
-version = "0.0.11"  # base version
+version = "0.0.13"  # base version
 packages = [{ include = "scorebook", from = "src" }]
 
 [tool.poetry.dependencies]
 python = ">=3.9,<3.14"
 datasets = ">=3.6.0"
 notebook = ">=7.4.5,<8.0.0"
-trismik = ">=1.0.1,<2.0.0"
+trismik = "1.0.2"
 ipywidgets = ">=8.0.0"
 
 # Optional dependencies
@@ -119,16 +119,22 @@ install_types = true
 [tool.flake8] # note that this depends on Flake8-pyproject
 ignore = ["D202", "W503", "W504"]
 
-[tool.unasync]
 [[tool.unasync.rules]]
 fromdir = "src/scorebook/evaluate/_async/"
 todir = "src/scorebook/evaluate/_sync/"
+replacements."scorebook.score._async.score_async" = "scorebook.score._sync.score"
+replacements."scorebook.score._async" = "scorebook.score._sync"
+replacements.evaluate_async = "evaluate"
+replacements."Asynchronous evaluation complete" = "Synchronous evaluation complete"
+replacements."    run_results = asyncio.gather(*[worker(run) for run in runs])" = "    run_results = [worker(run) for run in runs]"
+replacements.async_nullcontext = "nullcontext"
+replacements.create_trismik_async_client = "create_trismik_sync_client"
+replacements.score_async = "score"
 
-
-# Custom replacements beyond default async/await transformations
-[tool.unasync.rules.replacements]
-"evaluate_async" = "evaluate"
-"Asynchronous evaluation complete" = "Synchronous evaluation complete"
-"    run_results = asyncio.gather(*[worker(run) for run in runs])" = "    run_results = [worker(run) for run in runs]"
-"async_nullcontext" = "nullcontext"
-"create_trismik_async_client" = "create_trismik_sync_client"
+[[tool.unasync.rules]]
+fromdir = "src/scorebook/score/_async/"
+todir = "src/scorebook/score/_sync/"
+replacements.score_async = "score"
+replacements."Async scoring complete" = "Scoring complete"
+replacements.calculate_metric_scores_async = "calculate_metric_scores"
+replacements.upload_result_async = "upload_result"

{scorebook-0.0.11 → scorebook-0.0.13}/src/scorebook/__init__.py

@@ -12,15 +12,22 @@ __version__ = importlib.metadata.version(__package__ or __name__)
 from scorebook.eval_datasets import EvalDataset
 from scorebook.evaluate import evaluate, evaluate_async
 from scorebook.inference.inference_pipeline import InferencePipeline
-from scorebook.trismik.credentials import login, whoami
+from scorebook.score import score, score_async
+from scorebook.trismik.credentials import login, logout, whoami
+from scorebook.trismik.upload_results import upload_result, upload_result_async
 from scorebook.utils.render_template import render_template
 
 __all__ = [
     "EvalDataset",
     "evaluate",
     "evaluate_async",
+    "score",
+    "score_async",
     "render_template",
     "login",
+    "logout",
     "whoami",
     "InferencePipeline",
+    "upload_result",
+    "upload_result_async",
 ]

{scorebook-0.0.11 → scorebook-0.0.13}/src/scorebook/eval_datasets/eval_dataset.py

@@ -137,6 +137,24 @@ class EvalDataset:
             raise DatasetNotInitializedError("Dataset is not initialized")
         return list(map(str, self._hf_dataset.column_names))
 
+    @property
+    def split(self) -> Optional[str]:
+        """Return the split name of the underlying HuggingFace dataset, if available.
+
+        Returns:
+            The split name (e.g., "train", "test", "validation") if the dataset was loaded
+            from HuggingFace with a specific split. Returns None if the dataset was created
+            from a list, CSV, JSON, or loaded without a split specification.
+
+        Raises:
+            DatasetNotInitializedError: If the dataset is not initialized.
+        """
+        if self._hf_dataset is None:
+            raise DatasetNotInitializedError("Dataset is not initialized")
+
+        split = self._hf_dataset.split
+        return str(split) if split is not None else None
+
     def shuffle(self) -> None:
         """Randomly shuffle the dataset items."""
         if self._hf_dataset is None:

{scorebook-0.0.11 → scorebook-0.0.13}/src/scorebook/evaluate/_async/evaluate_async.py

@@ -3,13 +3,8 @@ import logging
 from typing import Any, Callable, Dict, List, Literal, Optional, Union, cast
 
 from trismik import TrismikAsyncClient, TrismikClient
-from trismik.types import (
-    TrismikClassicEvalItem,
-    TrismikClassicEvalMetric,
-    TrismikClassicEvalRequest,
-    TrismikClassicEvalResponse,
-    TrismikRunMetadata,
-)
+from trismik.settings import evaluation_settings
+from trismik.types import TrismikRunMetadata
 
 from scorebook.eval_datasets import EvalDataset
 from scorebook.evaluate.evaluate_helpers import (
@@ -20,12 +15,12 @@ from scorebook.evaluate.evaluate_helpers import (
     make_trismik_inference,
     prepare_datasets,
     prepare_hyperparameter_configs,
-    resolve_show_progress,
-    resolve_upload_results,
-    score_metrics,
+    resolve_adaptive_split,
     validate_parameters,
 )
 from scorebook.exceptions import InferenceError, ScoreBookError
+from scorebook.inference.inference_pipeline import InferencePipeline
+from scorebook.score._async.score_async import score_async
 from scorebook.types import (
     AdaptiveEvalRunResult,
     AdaptiveEvalRunSpec,
@@ -33,14 +28,20 @@ from scorebook.types import (
     EvalResult,
     EvalRunSpec,
 )
-from scorebook.utils import async_nullcontext, evaluation_progress_context
+from scorebook.utils import (
+    async_nullcontext,
+    evaluation_progress_context,
+    resolve_show_progress,
+    resolve_upload_results,
+)
 
 logger = logging.getLogger(__name__)
 
 
 async def evaluate_async(
-    inference: Callable,
+    inference: Union[Callable, InferencePipeline],
     datasets: Union[str, EvalDataset, List[Union[str, EvalDataset]]],
+    split: Optional[str] = None,
     hyperparameters: Optional[Union[Dict[str, Any], List[Dict[str, Any]]]] = None,
     metadata: Optional[Dict[str, Any]] = None,
     experiment_id: Optional[str] = None,
@@ -59,6 +60,7 @@ async def evaluate_async(
     Args:
         inference: The inference callable to evaluate
         datasets: Dataset(s) to evaluate on
+        split: Split to use for evaluation (default: "validation")
         hyperparameters: Hyperparameter configuration(s) to evaluate with
         metadata: Optional metadata to attach to the evaluation
         experiment_id: Optional experiment identifier
@@ -83,14 +85,14 @@ async def evaluate_async(
     validate_parameters(locals(), evaluate_async)
 
     # Prepare datasets, hyperparameters, and eval run specs
-    datasets = prepare_datasets(datasets, sample_size)
+    datasets = prepare_datasets(datasets, split, sample_size)
     hyperparameter_configs = prepare_hyperparameter_configs(hyperparameters)
     eval_run_specs = sorted(
         build_eval_run_specs(datasets, hyperparameter_configs, experiment_id, project_id, metadata),
         key=lambda run: (run.dataset_index, run.hyperparameters_index),
     )
 
-    # Create Trismik client if needed (for adaptive evals or uploads)
+    # Create a Trismik client if needed (for adaptive evals or uploads)
     needs_client = upload_results or any(
         isinstance(run, AdaptiveEvalRunSpec) for run in eval_run_specs
     )
@@ -101,7 +103,14 @@ async def evaluate_async(
     async with trismik_client or async_nullcontext():
         # Execute evaluation runs
         # Calculate total items across all runs
-        total_items = sum(len(run.dataset.items) for run in eval_run_specs)
+        total_items = sum(
+            (
+                len(run.dataset.items)
+                if isinstance(run, EvalRunSpec)
+                else evaluation_settings["max_iterations"]
+            )  # Adaptive evals use max_iterations
+            for run in eval_run_specs
+        )
         model_display = get_model_name(inference)
 
         with evaluation_progress_context(
@@ -145,34 +154,32 @@ async def execute_runs(
     async def worker(
         run: Union[EvalRunSpec, AdaptiveEvalRunSpec]
     ) -> Union[ClassicEvalRunResult, AdaptiveEvalRunResult]:
+        # Execute run (score_async handles upload internally for classic evals)
         run_result = await execute_run(
-            inference, run, experiment_id, project_id, metadata, trismik_client
+            inference, run, upload_results, experiment_id, project_id, metadata, trismik_client
         )
+
         # Update progress bars with items processed and success status
         if progress_bars is not None:
-            items_processed = len(run.dataset.items)
+            # Classic evals have .items; adaptive evals use max_iterations
+            items_processed = (
+                len(run.dataset.items)
+                if isinstance(run, EvalRunSpec)
+                else evaluation_settings["max_iterations"]
+            )
             progress_bars.on_run_completed(items_processed, run_result.run_completed)
 
+        # Update upload progress for classic evals
         if (
             upload_results
             and isinstance(run_result, ClassicEvalRunResult)
-            and experiment_id
-            and project_id
             and run_result.run_completed
-            and trismik_client is not None
         ):
-            try:
-                run_id = await upload_classic_run_results(
-                    run_result, experiment_id, project_id, inference, metadata, trismik_client
-                )
-                run_result.run_id = run_id
+            # Check if upload succeeded by checking for run_id
+            if experiment_id and project_id:
+                upload_succeeded = run_result.run_id is not None
                 if progress_bars is not None:
-                    progress_bars.on_upload_completed(succeeded=True)
-            except Exception as e:
-                logger.warning(f"Failed to upload run results: {e}")
-                if progress_bars is not None:
-                    progress_bars.on_upload_completed(succeeded=False)
-                # Continue evaluation even if upload fails
+                    progress_bars.on_upload_completed(succeeded=upload_succeeded)
 
         return run_result
 
@@ -191,6 +198,7 @@ async def execute_runs(
 async def execute_run(
     inference: Callable,
     run: Union[EvalRunSpec, AdaptiveEvalRunSpec],
+    upload_results: bool,  # NEW PARAMETER
     experiment_id: Optional[str] = None,
     project_id: Optional[str] = None,
     metadata: Optional[Dict[str, Any]] = None,
@@ -199,7 +207,9 @@ async def execute_run(
     """Execute a single evaluation run."""
 
     if isinstance(run, EvalRunSpec):
-        return await execute_classic_eval_run(inference, run)
+        return await execute_classic_eval_run(
+            inference, run, upload_results, experiment_id, project_id, metadata
+        )
 
     elif isinstance(run, AdaptiveEvalRunSpec):
         resolved_experiment_id = experiment_id if experiment_id is not None else run.experiment_id
@@ -217,24 +227,79 @@ async def execute_run(
         raise ScoreBookError(f"An internal error occurred: {type(run)} is not a valid run type")
 
 
-async def execute_classic_eval_run(inference: Callable, run: EvalRunSpec) -> ClassicEvalRunResult:
-    """Execute a classic evaluation run."""
+async def execute_classic_eval_run(
+    inference: Callable,
+    run: EvalRunSpec,
+    upload_results: bool,
+    experiment_id: Optional[str],
+    project_id: Optional[str],
+    metadata: Optional[Dict[str, Any]],
+) -> ClassicEvalRunResult:
+    """Execute a classic evaluation run using score_async() for scoring and uploading."""
     logger.debug("Executing classic eval run for %s", run)
 
     inference_outputs = None
-    metric_scores = None
+    scores = None
 
     try:
+        # 1. Run inference
         inference_outputs = await run_inference_callable(
             inference, run.inputs, run.hyperparameter_config
         )
-        metric_scores = score_metrics(run.dataset, inference_outputs, run.labels)
-        logger.debug("Classic evaluation completed for run %s", run)
-        return ClassicEvalRunResult(run, True, inference_outputs, metric_scores)
+
+        # 2. Build items for score_async
+        items = [
+            {
+                "input": run.inputs[i] if i < len(run.inputs) else None,
+                "output": inference_outputs[i],
+                "label": run.labels[i] if i < len(run.labels) else "",
+            }
+            for i in range(len(inference_outputs))
+        ]
+
+        # 3. Get the model name for upload
+        model_name = get_model_name(inference, metadata)
+
+        # 4. Call score_async
+        scores = await score_async(
+            items=items,
+            metrics=run.dataset.metrics,
+            output_column="output",  # Explicit parameter
+            label_column="label",  # Explicit parameter
+            input_column="input",  # Explicit parameter
+            hyperparameters=run.hyperparameter_config,
+            dataset_name=run.dataset.name,
+            model_name=model_name,
+            metadata=metadata,
+            experiment_id=experiment_id,
+            project_id=project_id,
+            upload_results=upload_results,
+            show_progress=False,
+        )
+
+        # 5. Extract run_id if upload succeeded
+        run_id = None
+        if scores.get("aggregate_results") and len(scores["aggregate_results"]) > 0:
+            run_id = scores["aggregate_results"][0].get("run_id")
+
+        logger.debug("Classic evaluation completed for run %s (run_id: %s)", run, run_id)
+        return ClassicEvalRunResult(
+            run_spec=run,
+            run_completed=True,
+            outputs=inference_outputs,
+            scores=scores,
+            run_id=run_id,
+        )
 
     except Exception as e:
         logger.warning("Failed to complete classic eval run for %s: %s", run, str(e))
-        return ClassicEvalRunResult(run, False, inference_outputs, metric_scores)
+        return ClassicEvalRunResult(
+            run_spec=run,
+            run_completed=False,
+            outputs=inference_outputs,
+            scores=scores,
+            run_id=None,
+        )
 
 
 async def run_inference_callable(
@@ -296,93 +361,6 @@ async def execute_adaptive_eval_run(
         return AdaptiveEvalRunResult(run, False, {})
 
 
-async def upload_classic_run_results(
-    run_result: ClassicEvalRunResult,
-    experiment_id: str,
-    project_id: str,
-    inference_callable: Optional[Callable],
-    metadata: Optional[Dict[str, Any]],
-    trismik_client: Union[TrismikClient, TrismikAsyncClient],
-) -> str:
-    """Upload a classic evaluation run result to Trismik platform.
-
-    Args:
-        run: The evaluation run result to upload
-        experiment_id: Trismik experiment identifier
-        project_id: Trismik project identifier
-        model: Model name used for evaluation
-        metadata: Optional metadata dictionary
-        trismik_client: Trismik client instance
-
-    Returns:
-        Run id
-    """
-    model = get_model_name(inference_callable)
-
-    # Create eval items from run_spec inputs, outputs, and labels
-    items: List[TrismikClassicEvalItem] = []
-    inputs_outputs = zip(run_result.run_spec.inputs, run_result.outputs)
-    for idx, (input_value, output) in enumerate(inputs_outputs):
-        labels = run_result.run_spec.labels
-        label = labels[idx] if idx < len(labels) else ""
-
-        # Calculate item-level metrics for this item
-        item_metrics: Dict[str, Any] = {}
-        if run_result.scores:
-            for metric_name, metric_data in run_result.scores.items():
-                if isinstance(metric_data, dict) and "item_scores" in metric_data:
-                    if idx < len(metric_data["item_scores"]):
-                        item_metrics[metric_name] = metric_data["item_scores"][idx]
-                else:
-                    # If scores is just a single value, use it for all items
-                    item_metrics[metric_name] = metric_data
-
-        eval_item = TrismikClassicEvalItem(
-            datasetItemId=str(idx),
-            modelInput=str(input_value),
-            modelOutput=str(output),
-            goldOutput=str(label),
-            metrics=item_metrics,
-        )
-        items.append(eval_item)
-
-    # Create eval metrics from run aggregate scores
-    metrics: List[TrismikClassicEvalMetric] = []
-    if run_result.scores:
-        for metric_name, metric_data in run_result.scores.items():
-            if isinstance(metric_data, dict) and "aggregate_scores" in metric_data:
-                # Handle structured metric data with aggregate scores
-                for agg_name, agg_value in metric_data["aggregate_scores"].items():
-                    metric_id = (
-                        f"{metric_name}_{agg_name}" if agg_name != metric_name else metric_name
-                    )
-                    metric = TrismikClassicEvalMetric(metricId=metric_id, value=agg_value)
-                    metrics.append(metric)
-            else:
-                # Handle simple metric data (single value)
-                metric = TrismikClassicEvalMetric(metricId=metric_name, value=metric_data)
-                metrics.append(metric)
-
-    classic_eval_request = TrismikClassicEvalRequest(
-        project_id,
-        experiment_id,
-        run_result.run_spec.dataset.name,
-        model,
-        run_result.run_spec.hyperparameter_config,
-        items,
-        metrics,
-    )
-
-    response: TrismikClassicEvalResponse = await trismik_client.submit_classic_eval(
-        classic_eval_request
-    )
-
-    run_id: str = response.id
-    logger.info(f"Classic eval run uploaded successfully with run_id: {run_id}")
-
-    return run_id
-
-
 async def run_adaptive_evaluation(
     inference: Callable,
     adaptive_run_spec: AdaptiveEvalRunSpec,
@@ -403,8 +381,20 @@ async def run_adaptive_evaluation(
     Returns:
         Results from the adaptive evaluation
     """
+    # Fetch available splits from Trismik
+    dataset_info = await trismik_client.get_dataset_info(adaptive_run_spec.dataset)
+    available_splits = dataset_info.splits if hasattr(dataset_info, "splits") else []
+
+    # Resolve the split to use (with fallback: user-specified -> validation -> test)
+    resolved_split = resolve_adaptive_split(
+        test_id=adaptive_run_spec.dataset,
+        user_specified_split=adaptive_run_spec.split,
+        available_splits=available_splits,
+    )
+
     trismik_results = await trismik_client.run(
         test_id=adaptive_run_spec.dataset,
+        split=resolved_split,
         project_id=project_id,
         experiment=experiment_id,
         run_metadata=TrismikRunMetadata(