sora 13274243a0 Bump vendored EvalScope and add K3-ready DPV4 configs.

Keep K3 suite selection and report-schema scoring in bash, merge K3/vision dataset_args into dpv4 yamls, and pin EvalScope at 735d920ee911 with local patches.

Co-authored-by: Cursor <cursoragent@cursor.com>

2026-09-02 07:30:48 +00:00

3.9 KiB

Raw Blame History

TriviaQA-Indic-MCQ

Overview

TriviaQA-Indic-MCQ reformats TriviaQA trivia questions as 4-way multiple-choice questions, translated into 10 Indic languages plus English, for evaluating multilingual world-knowledge recall.

Task Description

Task Type: Multilingual Multiple-Choice Trivia Question Answering
Input: Trivia question with 4 answer choices in one of 11 languages
Output: Correct answer letter
Languages: Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu

Evaluation Notes

Default configuration uses 0-shot evaluation (validation split, the only split available)
Use subset_list to evaluate specific languages (e.g., ['hi', 'ta']), or limit to cap sample count — the full default run is ~18k samples per language across all 11 languages (~198k total)

Properties

Property	Value
Benchmark Name	`triviaqa_indic`
Dataset ID	sarvamai/trivia-qa-indic-mcq
Paper	N/A
Tags	`Knowledge`, `MCQ`, `MultiLingual`
Metrics	`accuracy`
Default Shots	0-shot
Evaluation Split	`validation`

Data Statistics

Metric	Value
Total Samples	197,384
Prompt Length (Mean)	353.95 chars
Prompt Length (Min/Max)	247 / 1267 chars

Per-Subset Statistics:

Subset	Samples	Prompt Mean	Prompt Min	Prompt Max
`bn`	17,944	351.81	253	1131
`en`	17,944	347.69	256	1158
`gu`	17,944	347.73	247	1117
`hi`	17,944	349.05	258	1141
`kn`	17,944	359.89	253	1164
`ml`	17,944	366.11	257	1267
`mr`	17,944	350.9	253	1078
`or`	17,944	349.44	255	941
`pa`	17,944	345.48	253	1157
`ta`	17,944	368.55	248	1198
`te`	17,944	356.75	254	1259

Sample Example

Subset: bn

{
  "input": [
    {
      "id": "eff7630c",
      "content": "Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of A,B,C,D.\n\nচিপমঙ্কসের পিছনে লোকটি কে ছিল?\n\nA) ডেভিড সেভিল\nB) জাগরেব শহর - ক্রোয়েশিয়া প্রজাতন্ত্র\nC) পবিত্র ক্রুসেড\nD) উপাদান (অ্যালবাম)"
    }
  ],
  "choices": [
    "ডেভিড সেভিল",
    "জাগরেব শহর - ক্রোয়েশিয়া প্রজাতন্ত্র",
    "পবিত্র ক্রুসেড",
    "উপাদান (অ্যালবাম)"
  ],
  "target": "A",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "language": "Bengali"
  }
}

Prompt Template

Prompt Template:

Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.

{question}

{choices}

Usage

Using CLI

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets triviaqa_indic \
    --limit 10  # Remove this line for formal evaluation

Using Python

from evalscope import run_task
from evalscope.config import TaskConfig

task_cfg = TaskConfig(
    model='YOUR_MODEL',
    api_url='OPENAI_API_COMPAT_URL',
    api_key='EMPTY_TOKEN',
    datasets=['triviaqa_indic'],
    dataset_args={
        'triviaqa_indic': {
            # subset_list: ['bn', 'en', 'gu']  # optional, evaluate specific subsets
        }
    },
    limit=10,  # Remove this line for formal evaluation
)

run_task(task_cfg=task_cfg)

3.9 KiB Raw Blame History