guidellm 0.4.0a155__tar.gz → 0.4.0a173__tar.gz

{guidellm-0.4.0a155/src/guidellm.egg-info → guidellm-0.4.0a173}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: guidellm
-Version: 0.4.0a155
+Version: 0.4.0a173
 Summary: Guidance platform for deploying and managing large language models.
 Author: Red Hat
 License: Apache-2.0

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/__main__.py

@@ -156,8 +156,9 @@ def benchmark():
 )
 @click.option(
     "--rate",
-    type=float,
-    multiple=True,
+    type=str,
+    callback=cli_tools.parse_list_floats,
+    multiple=False,
     default=BenchmarkGenerativeTextArgs.get_default("rate"),
     help=(
         "Benchmark rate(s) to test. Meaning depends on profile: "
@@ -383,7 +384,7 @@ def run(**kwargs):
         kwargs.get("data_args"), default=[], simplify_single=False
     )
     kwargs["rate"] = cli_tools.format_list_arg(
-        kwargs.get("rate"), default=None, simplify_single=True
+        kwargs.get("rate"), default=None, simplify_single=False
     )
 
     disable_console_outputs = kwargs.pop("disable_console_outputs", False)

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/benchmark/benchmarker.py

@@ -57,6 +57,7 @@ class Benchmarker(
         backend: BackendInterface[RequestT, ResponseT],
         profile: Profile,
         environment: Environment,
+        data: list[Any],
         progress: BenchmarkerProgress[BenchmarkT] | None = None,
         sample_requests: int | None = 20,
         warmup: float | None = None,
@@ -149,6 +150,7 @@ class Benchmarker(
                     environment=environment,
                     strategy=strategy,
                     constraints=constraints,
+                    data=data,
                 )
                 if progress:
                     await progress.on_benchmark_complete(benchmark)

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/benchmark/entrypoints.py

@@ -436,6 +436,7 @@ async def benchmark_generative_text(
         backend=backend,
         profile=profile,
         environment=NonDistributedEnvironment(),
+        data=args.data,
         progress=progress,
         sample_requests=args.sample_requests,
         warmup=args.warmup,

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/benchmark/output.py

@@ -649,6 +649,8 @@ class GenerativeBenchmarkerCSV(GenerativeBenchmarkerOutput):
         status_dist_summary: StatusDistributionSummary = getattr(
             benchmark.metrics, metric
         )
+        if not hasattr(status_dist_summary, status):
+            return [], []
         dist_summary: DistributionSummary = getattr(status_dist_summary, status)
 
         headers = [
@@ -688,7 +690,7 @@ class GenerativeBenchmarkerCSV(GenerativeBenchmarkerOutput):
         values: list[str] = [
             benchmark.benchmarker.profile.model_dump_json(),
             json.dumps(benchmark.benchmarker.backend),
-            json.dumps(benchmark.benchmarker.requests["attributes"]["data"]),
+            json.dumps(benchmark.benchmarker.requests["data"]),
         ]
 
         if len(headers) != len(values):

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/benchmark/schemas.py

@@ -1674,6 +1674,7 @@ class GenerativeBenchmark(Benchmark, StandardBaseDict):
         environment: Environment,
         strategy: SchedulingStrategy,
         constraints: dict[str, dict[str, Any]],
+        data: list[Any],
     ) -> GenerativeBenchmark:
         """
         Compile final generative benchmark from accumulated state.
@@ -1702,7 +1703,7 @@ class GenerativeBenchmark(Benchmark, StandardBaseDict):
             ),
             benchmarker=BenchmarkerDict(
                 profile=profile,
-                requests=InfoMixin.extract_from_obj(requests),
+                requests={"data": data},
                 backend=backend.info,
                 environment=environment.info,
             ),

guidellm-0.4.0a173/src/guidellm/data/deserializers/deserializer.py

@@ -0,0 +1,144 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from typing import Any, Protocol, Union, runtime_checkable
+
+from datasets import Dataset, DatasetDict, IterableDataset, IterableDatasetDict
+from transformers import PreTrainedTokenizerBase
+
+from guidellm.data.utils import resolve_dataset_split
+from guidellm.utils import RegistryMixin
+
+__all__ = [
+    "DataNotSupportedError",
+    "DatasetDeserializer",
+    "DatasetDeserializerFactory",
+]
+
+
+class DataNotSupportedError(Exception):
+    """Exception raised when data format is not supported by deserializer."""
+
+
+@runtime_checkable
+class DatasetDeserializer(Protocol):
+    def __call__(
+        self,
+        data: Any,
+        processor_factory: Callable[[], PreTrainedTokenizerBase],
+        random_seed: int,
+        **data_kwargs: dict[str, Any],
+    ) -> Dataset | IterableDataset | DatasetDict | IterableDatasetDict: ...
+
+
+class DatasetDeserializerFactory(
+    RegistryMixin[Union["type[DatasetDeserializer]", DatasetDeserializer]],
+):
+    @classmethod
+    def deserialize(
+        cls,
+        data: Any,
+        processor_factory: Callable[[], PreTrainedTokenizerBase],
+        random_seed: int = 42,
+        type_: str | None = None,
+        resolve_split: bool = True,
+        select_columns: list[str] | None = None,
+        remove_columns: list[str] | None = None,
+        **data_kwargs: dict[str, Any],
+    ) -> Dataset | IterableDataset:
+        dataset: Dataset
+
+        if type_ is None:
+            dataset = cls._deserialize_with_registered_deserializers(
+                data, processor_factory, random_seed, **data_kwargs
+            )
+
+        else:
+            dataset = cls._deserialize_with_specified_deserializer(
+                data, type_, processor_factory, random_seed, **data_kwargs
+            )
+
+        if resolve_split:
+            dataset = resolve_dataset_split(dataset)
+
+        if select_columns is not None or remove_columns is not None:
+            column_names = dataset.column_names or list(next(iter(dataset)).keys())
+            if select_columns is not None:
+                remove_columns = [
+                    col for col in column_names if col not in select_columns
+                ]
+
+            dataset = dataset.remove_columns(remove_columns)
+
+        return dataset
+
+    @classmethod
+    def _deserialize_with_registered_deserializers(
+        cls,
+        data: Any,
+        processor_factory: Callable[[], PreTrainedTokenizerBase],
+        random_seed: int = 42,
+        **data_kwargs: dict[str, Any],
+    ) -> Dataset:
+        if cls.registry is None:
+            raise RuntimeError("registry is None; cannot deserialize dataset")
+        dataset: Dataset | None = None
+
+        errors: dict[str, Exception] = {}
+        # Note: There is no priority order for the deserializers, so all deserializers
+        #  must be mutually exclusive to ensure deterministic behavior.
+        for _name, deserializer in cls.registry.items():
+            deserializer_fn: DatasetDeserializer = (
+                deserializer() if isinstance(deserializer, type) else deserializer
+            )
+
+            try:
+                dataset = deserializer_fn(
+                    data=data,
+                    processor_factory=processor_factory,
+                    random_seed=random_seed,
+                    **data_kwargs,
+                )
+            except Exception as e:  # noqa: BLE001 # The exceptions are saved.
+                errors[_name] = e
+
+            if dataset is not None:
+                return dataset  # Success
+
+        if len(errors) > 0:
+            err_msgs = ""
+            def sort_key(item):
+                return (isinstance(item[1], DataNotSupportedError), item[0])
+            for key, err in sorted(errors.items(), key=sort_key):
+                err_msgs += f"\n  - Deserializer '{key}': ({type(err).__name__}) {err}"
+            raise ValueError(
+                "Data deserialization failed, likely because the input doesn't "
+                f"match any of the input formats. See the {len(errors)} error(s) that "
+                f"occurred while attempting to deserialize the data {data}:{err_msgs}"
+            )
+        return dataset
+
+    @classmethod
+    def _deserialize_with_specified_deserializer(
+        cls,
+        data: Any,
+        type_: str,
+        processor_factory: Callable[[], PreTrainedTokenizerBase],
+        random_seed: int = 42,
+        **data_kwargs: dict[str, Any],
+    ) -> Dataset:
+        deserializer_from_type = cls.get_registered_object(type_)
+        if deserializer_from_type is None:
+            raise ValueError(f"Deserializer type '{type_}' is not registered.")
+        if isinstance(deserializer_from_type, type):
+            deserializer_fn = deserializer_from_type()
+        else:
+            deserializer_fn = deserializer_from_type
+
+        return deserializer_fn(
+            data=data,
+            processor_factory=processor_factory,
+            random_seed=random_seed,
+            **data_kwargs,
+        )
+

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/deserializers/file.py

@@ -34,11 +34,11 @@ class TextFileDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)  # Ignore unused args format errors
 
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() not in {".txt", ".text"}
@@ -62,10 +62,10 @@ class CSVFileDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() != ".csv"
@@ -86,10 +86,10 @@ class JSONFileDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() not in {".json", ".jsonl"}
@@ -110,10 +110,10 @@ class ParquetFileDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() != ".parquet"
@@ -134,10 +134,10 @@ class ArrowFileDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() != ".arrow"
@@ -158,10 +158,10 @@ class HDF5FileDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() not in {".hdf5", ".h5"}
@@ -185,7 +185,7 @@ class DBFileDatasetDeserializer(DatasetDeserializer):
     ) -> dict[str, list]:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() != ".db"
@@ -209,7 +209,7 @@ class TarFileDatasetDeserializer(DatasetDeserializer):
     ) -> dict[str, list]:
         _ = (processor_factory, random_seed)
         if (
-            not isinstance(data, (str, Path))
+            not isinstance(data, str | Path)
             or not (path := Path(data)).exists()
             or not path.is_file()
             or path.suffix.lower() != ".tar"

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/deserializers/huggingface.py

@@ -36,7 +36,7 @@ class HuggingFaceDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset | IterableDataset | DatasetDict | IterableDatasetDict:
         _ = (processor_factory, random_seed)
 
         if isinstance(

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/deserializers/memory.py

@@ -33,7 +33,7 @@ class InMemoryDictDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)  # Ignore unused args format errors
 
         if (
@@ -67,7 +67,7 @@ class InMemoryDictListDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)  # Ignore unused args format errors
 
         if (
@@ -81,9 +81,9 @@ class InMemoryDictListDatasetDeserializer(DatasetDeserializer):
                 f"expected list of dicts, got {data}"
             )
 
-        data: list[dict[str, Any]] = cast("list[dict[str, Any]]", data)
-        first_keys = set(data[0].keys())
-        for index, item in enumerate(data):
+        typed_data: list[dict[str, Any]] = cast("list[dict[str, Any]]", data)
+        first_keys = set(typed_data[0].keys())
+        for index, item in enumerate(typed_data):
             if set(item.keys()) != first_keys:
                 raise DataNotSupportedError(
                     f"All dictionaries must have the same keys. "
@@ -92,8 +92,8 @@ class InMemoryDictListDatasetDeserializer(DatasetDeserializer):
                 )
 
         # Convert list of dicts to dict of lists
-        result_dict = {key: [] for key in first_keys}
-        for item in data:
+        result_dict: dict = {key: [] for key in first_keys}
+        for item in typed_data:
             for key, value in item.items():
                 result_dict[key].append(value)
 
@@ -108,7 +108,7 @@ class InMemoryItemListDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         _ = (processor_factory, random_seed)  # Ignore unused args format errors
 
         primitive_types = (str, int, float, bool, type(None))
@@ -135,7 +135,7 @@ class InMemoryJsonStrDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         if (
             isinstance(data, str)
             and (json_str := data.strip())
@@ -145,16 +145,18 @@ class InMemoryJsonStrDatasetDeserializer(DatasetDeserializer):
             )
         ):
             with contextlib.suppress(Exception):
-                parsed = json.loads(data)
+                parsed_data = json.loads(data)
 
-            for deserializer in [
-                InMemoryDictDatasetDeserializer,
-                InMemoryDictListDatasetDeserializer,
-                InMemoryItemListDatasetDeserializer,
-            ]:
+            deserializers = [
+                InMemoryDictDatasetDeserializer(),
+                InMemoryDictListDatasetDeserializer(),
+                InMemoryItemListDatasetDeserializer(),
+            ]
+
+            for deserializer in deserializers:
                 with contextlib.suppress(DataNotSupportedError):
-                    return deserializer()(
-                        parsed, data_kwargs, processor_factory, random_seed
+                    return deserializer(
+                        parsed_data, processor_factory, random_seed, **data_kwargs
                     )
 
         raise DataNotSupportedError(
@@ -171,7 +173,7 @@ class InMemoryCsvDatasetDeserializer(DatasetDeserializer):
         processor_factory: Callable[[], PreTrainedTokenizerBase],
         random_seed: int,
         **data_kwargs: dict[str, Any],
-    ) -> dict[str, list]:
+    ) -> Dataset:
         if (
             isinstance(data, str)
             and (csv_str := data.strip())

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/deserializers/synthetic.py

@@ -99,21 +99,23 @@ class SyntheticTextDatasetConfig(StandardBaseModel):
 
     @model_validator(mode="after")
     def check_prefix_options(self) -> SyntheticTextDatasetConfig:
-        prefix_count = self.__pydantic_extra__.get("prefix_count", None)  # type: ignore[attr-defined]
-        prefix_tokens = self.__pydantic_extra__.get("prefix_tokens", None)  # type: ignore[attr-defined]
-        if prefix_count is not None or prefix_tokens is not None:
-            if self.prefix_buckets:
-                raise ValueError(
-                    "prefix_buckets is mutually exclusive"
-                    " with prefix_count and prefix_tokens"
-                )
+        if self.__pydantic_extra__ is not None:
+            prefix_count = self.__pydantic_extra__.get("prefix_count", None)  # type: ignore[attr-defined]
+            prefix_tokens = self.__pydantic_extra__.get("prefix_tokens", None)  # type: ignore[attr-defined]
+
+            if prefix_count is not None or prefix_tokens is not None:
+                if self.prefix_buckets:
+                    raise ValueError(
+                        "prefix_buckets is mutually exclusive"
+                        " with prefix_count and prefix_tokens"
+                    )
 
-            self.prefix_buckets = [
-                SyntheticTextPrefixBucketConfig(
-                    prefix_count=prefix_count or 1,
-                    prefix_tokens=prefix_tokens or 0,
-                )
-            ]
+                self.prefix_buckets = [
+                    SyntheticTextPrefixBucketConfig(
+                        prefix_count=prefix_count or 1,
+                        prefix_tokens=prefix_tokens or 0,
+                    )
+                ]
 
         return self
 
@@ -174,14 +176,14 @@ class SyntheticTextGenerator:
     def _create_prompt(
         self, prompt_tokens_count: int, faker: Faker, unique: str = ""
     ) -> str:
-        prompt_token_ids = []
+        prompt_token_ids: list[int] = []
         avg_chars_per_token = 5
         margin_of_safety = 1.5
         attempts = 0
 
         while len(prompt_token_ids) < prompt_tokens_count:
             attempts += 1
-            num_chars = (
+            num_chars = int(
                 prompt_tokens_count * avg_chars_per_token * margin_of_safety * attempts
             )
             text = unique + faker.text(max_nb_chars=num_chars)

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/loaders.py

@@ -17,6 +17,7 @@ from guidellm.logger import logger
 __all__ = ["DataLoader", "DatasetsIterator"]
 
 
+
 class DatasetsIterator(TorchIterableDataset):
     def __init__(
         self,
@@ -85,7 +86,7 @@ class DatasetsIterator(TorchIterableDataset):
 
             while max_items is None or gen_count < max_items:
                 try:
-                    row = {
+                    row: dict[str, Any] = {
                         "items": [next(dataset_iter) for dataset_iter in dataset_iters]
                     }
                     gen_count += 1
@@ -98,9 +99,12 @@ class DatasetsIterator(TorchIterableDataset):
                         continue
 
                     for preprocessor in self.preprocessors:
-                        row = preprocessor(row)
+                        # This can assign a GenerationRequest, which would then be
+                        # passed into the preprocessor, which is a type violation.
+                        # This should be fixed at some point.
+                        row = preprocessor(row)  # type: ignore[assignment]
                     yield row
-                except Exception as err:
+                except Exception as err:  # noqa: BLE001 # Exception logged
                     logger.error(f"Skipping data row due to error: {err}")
                     gen_count -= 1
 

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/preprocessors/formatters.py

@@ -7,8 +7,6 @@ from guidellm.data.preprocessors.preprocessor import (
     DatasetPreprocessor,
     PreprocessorRegistry,
 )
-from guidellm.data.schemas import GenerativeDatasetColumnType
-from guidellm.data.utils import text_stats
 from guidellm.schemas import GenerationRequest, GenerationRequestArguments, UsageMetrics
 
 __all__ = [
@@ -59,9 +57,13 @@ class GenerativeTextCompletionsRequestFormatter(RequestFormatter):
         self.max_tokens: int | None = max_tokens or max_completion_tokens
 
     def __call__(
-        self, columns: dict[GenerativeDatasetColumnType, list[Any]]
+        self, columns: dict[str, list[Any]]
     ) -> GenerationRequest:
-        arguments: GenerationRequestArguments = GenerationRequestArguments(body={})
+        """
+        :param columns: A dict of GenerativeDatasetColumnType to Any
+        """
+        arguments: GenerationRequestArguments = GenerationRequestArguments()
+        arguments.body = {}  # The type checker works better setting this field here
         input_metrics = UsageMetrics()
         output_metrics = UsageMetrics()
 
@@ -99,10 +101,9 @@ class GenerativeTextCompletionsRequestFormatter(RequestFormatter):
         prefix = "".join(pre for pre in columns.get("prefix_column", []) if pre)
         text = "".join(txt for txt in columns.get("text_column", []) if txt)
         if prefix or text:
-            arguments.body["prompt"] = prefix + text
-            stats = text_stats(arguments.body["prompt"])
-            input_metrics.text_characters = stats.get("num_chars")
-            input_metrics.text_words = stats.get("num_words")
+            prompt = prefix + text
+            arguments.body["prompt"] = prompt
+            input_metrics.add_text_metrics(prompt)
 
         return GenerationRequest(
             request_type="text_completions",
@@ -142,9 +143,13 @@ class GenerativeChatCompletionsRequestFormatter(RequestFormatter):
         )
 
     def __call__(  # noqa: C901, PLR0912, PLR0915
-        self, columns: dict[GenerativeDatasetColumnType, list[Any]]
+        self, columns: dict[str, list[Any]]
     ) -> GenerationRequest:
-        arguments = GenerationRequestArguments(body={})
+        """
+        :param columns: A dict of GenerativeDatasetColumnType to Any
+        """
+        arguments = GenerationRequestArguments()
+        arguments.body = {}  # The type checker works best with body assigned here
         input_metrics = UsageMetrics()
         output_metrics = UsageMetrics()
 
@@ -191,27 +196,14 @@ class GenerativeChatCompletionsRequestFormatter(RequestFormatter):
             if not prefix:
                 continue
 
-            stats = text_stats(prefix)
-            if (num_chars := stats.get("num_chars")) is not None:
-                input_metrics.text_characters = (
-                    input_metrics.text_characters or 0
-                ) + num_chars
-            if (num_words := stats.get("num_words")) is not None:
-                input_metrics.text_words = (input_metrics.text_words or 0) + num_words
-
+            input_metrics.add_text_metrics(prefix)
             arguments.body["messages"].append({"role": "system", "content": prefix})
 
         for text in columns.get("text_column", []):
             if not text:
                 continue
 
-            stats = text_stats(text)
-            if (num_chars := stats.get("num_chars")) is not None:
-                input_metrics.text_characters = (
-                    input_metrics.text_characters or 0
-                ) + num_chars
-            if (num_words := stats.get("num_words")) is not None:
-                input_metrics.text_words = (input_metrics.text_words or 0) + num_words
+            input_metrics.add_text_metrics(text)
 
             arguments.body["messages"].append(
                 {"role": "user", "content": [{"type": "text", "text": text}]}
@@ -329,9 +321,10 @@ class GenerativeAudioTranscriptionRequestFormatter(RequestFormatter):
         self.encode_audio_kwargs = encode_kwargs or {}
 
     def __call__(  # noqa: C901
-        self, columns: dict[GenerativeDatasetColumnType, list[Any]]
+        self, columns: dict[str, list[Any]]
     ) -> GenerationRequest:
-        arguments = GenerationRequestArguments(body={}, files={})
+        arguments = GenerationRequestArguments(files={})
+        arguments.body = {}  # The type checker works best with body assigned here
         input_metrics = UsageMetrics()
         output_metrics = UsageMetrics()
 
@@ -387,10 +380,9 @@ class GenerativeAudioTranscriptionRequestFormatter(RequestFormatter):
         prefix = "".join(pre for pre in columns.get("prefix_column", []) if pre)
         text = "".join(txt for txt in columns.get("text_column", []) if txt)
         if prefix or text:
-            arguments.body["prompt"] = prefix + text
-            stats = text_stats(arguments.body["prompt"])
-            input_metrics.text_characters = stats.get("num_chars")
-            input_metrics.text_words = stats.get("num_words")
+            prompt = prefix + text
+            arguments.body["prompt"] = prompt
+            input_metrics.add_text_metrics(prompt)
 
         return GenerationRequest(
             request_type="audio_transcriptions",
@@ -405,7 +397,7 @@ class GenerativeAudioTranslationRequestFormatter(
     GenerativeAudioTranscriptionRequestFormatter
 ):
     def __call__(
-        self, columns: dict[GenerativeDatasetColumnType, list[Any]]
+        self, columns: dict[str, list[Any]]
     ) -> GenerationRequest:
         result = super().__call__(columns)
         result.request_type = "audio_translations"

{guidellm-0.4.0a155 → guidellm-0.4.0a173}/src/guidellm/data/preprocessors/mappers.py

@@ -169,12 +169,12 @@ class GenerativeColumnMapper(DataDependentPreprocessor):
 
     def __call__(
         self, row: dict[str, Any]
-    ) -> dict[GenerativeDatasetColumnType, list[Any]]:
+    ) -> dict[str, list[Any]]:
         if self.datasets_column_mappings is None:
             raise ValueError("DefaultGenerativeColumnMapper not setup with data.")
 
         items = cast("dict[int, dict[str, Any]]", row.pop("items"))
-        mapped: dict[GenerativeDatasetColumnType, list[Any]] = defaultdict(list)
+        mapped: dict[str, Any] = defaultdict(list)
 
         for column_type, column_mappings in self.datasets_column_mappings.items():
             for (