Hugging Face's logo

Datasets:
jmrenouard
/
microllm-multilingual-benchmarks

Tasks:
Text Classification
Zero-Shot Classification
Sub-tasks:
multi-class-classification
natural-language-inference
Languages:
English
French
Spanish
Size:
n<1K
Tags:
microllm
benchmark
multilingual
customer-support
intent-classification
routing
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
id: string
language: string
lang: string
domain: string
track: string
question_type: string
decision_type: string
state: string
message: string
text: string
category: string
instructions: string
criteria: struct<billing: string, shipping: string, tech: string>
  child 0, billing: string
  child 1, shipping: string
  child 2, tech: string
rules: struct<billing: string, shipping: string, tech: string>
  child 0, billing: string
  child 1, shipping: string
  child 2, tech: string
options: struct<billing: string, shipping: string, tech: string>
  child 0, billing: string
  child 1, shipping: string
  child 2, tech: string
expected_answer: string
expected_result: string
metadata: struct<parallel_id: string, intent: string, languages_available: list<item: string>, source: string, (... 38 chars omitted)
  child 0, parallel_id: string
  child 1, intent: string
  child 2, languages_available: list<item: string>
      child 0, item: string
  child 3, source: string
  child 4, strategy_targets: list<item: string>
      child 0, item: string
to
{'id': Value('string'), 'language': Value('string'), 'domain': Value('string'), 'state': Value('string'), 'question_type': Value('string'), 'criteria': {'billing': Value('string'), 'shipping': Value('string'), 'tech': Value('string')}, 'expected_answer': Value('string'), 'metadata': {'parallel_id': Value('string'), 'intent': Value('string')}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              id: string
              language: string
              lang: string
              domain: string
              track: string
              question_type: string
              decision_type: string
              state: string
              message: string
              text: string
              category: string
              instructions: string
              criteria: struct<billing: string, shipping: string, tech: string>
                child 0, billing: string
                child 1, shipping: string
                child 2, tech: string
              rules: struct<billing: string, shipping: string, tech: string>
                child 0, billing: string
                child 1, shipping: string
                child 2, tech: string
              options: struct<billing: string, shipping: string, tech: string>
                child 0, billing: string
                child 1, shipping: string
                child 2, tech: string
              expected_answer: string
              expected_result: string
              metadata: struct<parallel_id: string, intent: string, languages_available: list<item: string>, source: string, (... 38 chars omitted)
                child 0, parallel_id: string
                child 1, intent: string
                child 2, languages_available: list<item: string>
                    child 0, item: string
                child 3, source: string
                child 4, strategy_targets: list<item: string>
                    child 0, item: string
              to
              {'id': Value('string'), 'language': Value('string'), 'domain': Value('string'), 'state': Value('string'), 'question_type': Value('string'), 'criteria': {'billing': Value('string'), 'shipping': Value('string'), 'tech': Value('string')}, 'expected_answer': Value('string'), 'metadata': {'parallel_id': Value('string'), 'intent': Value('string')}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:The task_ids "customer-support-intent-classification" is not in the official list: acceptability-classification, entity-linking-classification, fact-checking, intent-classification, language-identification, multi-class-classification, multi-label-classification, multi-input-text-classification, natural-language-inference, semantic-similarity-classification, sentiment-classification, topic-classification, semantic-similarity-scoring, sentiment-scoring, sentiment-analysis, hate-speech-detection, text-scoring, named-entity-recognition, part-of-speech, parsing, lemmatization, word-sense-disambiguation, coreference-resolution, extractive-qa, open-domain-qa, closed-domain-qa, news-articles-summarization, news-articles-headline-generation, dialogue-modeling, dialogue-generation, conversational, language-modeling, text-simplification, explanation-generation, abstractive-qa, open-domain-abstractive-qa, closed-domain-qa, open-book-qa, closed-book-qa, text2text-generation, slot-filling, masked-language-modeling, keyword-spotting, speaker-identification, audio-intent-classification, audio-emotion-recognition, audio-language-identification, multi-label-image-classification, multi-class-image-classification, face-detection, vehicle-detection, instance-segmentation, semantic-segmentation, panoptic-segmentation, image-captioning, image-inpainting, image-colorization, super-resolution, grasping, task-planning, tabular-multi-class-classification, tabular-multi-label-classification, tabular-single-column-regression, rdf-to-text, multiple-choice-qa, multiple-choice-coreference-resolution, document-retrieval, utterance-retrieval, entity-linking-retrieval, fact-checking-retrieval, univariate-time-series-forecasting, multivariate-time-series-forecasting, visual-question-answering, document-question-answering, pose-estimation

MicroLLM Multilingual Customer Support Benchmarks (EN, FR, ES, PT)

Dataset Description

The MicroLLM Multilingual Customer Support Benchmark provides a rigorously aligned, parallel quadrilingual evaluation corpus designed to assess intent classification, ticket triage, and zero-shot decision-making across English (EN), French (FR), Spanish (ES), and Portuguese (PT).

Created for the MicroLLM project, this dataset eliminates cross-lingual translation drift through parallel probes (cs-001 through cs-012), ensuring identical underlying semantics, decision boundaries, and ticket complexity across all four supported languages.

Supported Languages and Split Structure

The corpus comprises 48 curated customer support tickets distributed symmetrically across 4 languages (12 tickets per language) and 3 primary operational categories:

  • Billing & Refunds (billing, 6 tickets per language): Double billing, unexpected fee inquiries, annual subscription rate hikes, billing address modifications, cancelled account charges, and hardware refund requests.
  • Shipping & Logistics (shipping, 3 tickets per language): International shipment tracking delays, packages marked delivered but missing, and crushed or damaged parcels.
  • Technical Support (tech, 3 tickets per language): API HTTP 500 authentication failures, database latency spikes, and mobile application crash loops.

Directory and Split Layout

  • data/microllm_multilingual_corpus.jsonl: Complete 48-ticket multilingual dataset.
  • data/languages/en.jsonl / splits/en.jsonl: 12 English evaluation tickets.
  • data/languages/fr.jsonl / splits/fr.jsonl: 12 French evaluation tickets.
  • data/languages/es.jsonl / splits/es.jsonl: 12 Spanish evaluation tickets.
  • data/languages/pt.jsonl / splits/pt.jsonl: 12 Portuguese evaluation tickets.
  • data/multilingual_benchmarks_results.json: Official benchmark telemetry and latency metrics.

Benchmark Evaluation Results

Official MicroLLM execution across 240 benchmark evaluations (48 probe tickets evaluated across 5 system configurations):

Strategy Configuration Engine / Model Overall Accuracy p50 Latency (ms) p95 Latency (ms) Cascade Rate
fast Strategy 1 Lexicon 100.0% 0.357 ms 0.812 ms 0.0%
neural_minilm Strategy 2 MiniLMv2 INT8 100.0% 0.364 ms 0.795 ms 0.0%
neural_mdeberta Strategy 2 mDeBERTa INT8 100.0% 0.143 ms 0.421 ms 0.0%
auto_minilm Auto-Cascade (Fast + MiniLM) 100.0% 0.102 ms 0.380 ms 10.42%
auto_mdeberta Auto-Cascade (Fast + mDeBERTa) 100.0% 0.102 ms 0.380 ms 10.42%

All configurations achieve 100.0% classification accuracy across all 4 languages, substantially outperforming the 33.3% random baseline while maintaining deterministic sub-millisecond inference on standard CPU hardware.

Quickstart and Usage

Loading via Hugging Face Datasets 

from datasets import load_dataset

# Load complete multilingual corpus
dataset = load_dataset("jmrenouard/microllm-multilingual-benchmarks", split="full")
print(f"Loaded {len(dataset)} multilingual tickets")

# Load French split only
fr_dataset = load_dataset("jmrenouard/microllm-multilingual-benchmarks", split="fr")
print(f"Loaded {len(fr_dataset)} French tickets")

Loading via MicroLLM Python SDK 

import json
from pathlib import Path

# Parse directly from staged or local corpus
with open("benchmarks/microllm_multilingual_corpus.jsonl", "r", encoding="utf-8") as f:
    tickets = [json.loads(line) for line in f]

print(f"Validated {len(tickets)} customer tickets across EN, FR, ES, PT.")

Schema Specification

Each record in the dataset conforms to the following schema:

{
  "id": "mling-cs-001-fr",
  "language": "fr",
  "domain": "customer_support",
  "state": "Bonjour, j'ai ete facture deux fois ce mois-ci pour mon abonnement. Pouvez-vous me rembourser le deuxieme prelevement ?",
  "question_type": "choice",
  "criteria": {
    "billing": "Demandes relatives aux factures, paiements, prelevements et remboursements.",
    "shipping": "Questions concernant l'expedition, la livraison et le suivi de colis.",
    "tech": "Problemes techniques, pannes, erreurs logicielles et bogues."
  },
  "expected_answer": "billing",
  "metadata": {
    "parallel_id": "cs-001",
    "intent": "double_charge_refund"
  }
}

Citations and References

If you use this benchmark in academic or engineering publications, please cite:

@dataset{microllm_multilingual_benchmarks_2026,
  author    = {Renouard, Jean-Marie and MicroLLM Contributors},
  title     = {MicroLLM Multilingual Customer Support Benchmarks: Parallel EN, FR, ES, PT Evaluation Corpus},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/jmrenouard/microllm-multilingual-benchmarks}
}
Downloads last month
24
Total file size:
501 kB