datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
medmcqa-openai-native
MedMCQA — OpenAI-native, with a usable test split
MedMCQA is one of the most downloaded medical QA datasets on the Hub. Its test split has been unusable since release: all 6,150 rows carry cop=-1 (no label) and an empty explanation. You cannot score a model on it.
This release rebuilds a labelled, leak-free test split and converts everything to the native messages format, so it loads straight into TRL with no custom parsing.
What was actually wrong
Measured on the… See the full description on the dataset page: https://huggingface.co/datasets/Archangel-system/medmcqa-openai-native.medmcqa_ko_translatedMedMCQA
MedMCQA-CoT: 医学多肢選択問題with Chain-of-Thought推論
データセット概要
MedMCQA-CoTは、MedMCQAデータセットの拡張版で、各医学多肢選択問題に高品質なChain-of-Thought（CoT）推論を追加したデータセットです。医学的な推論プロセスを説明できるAIシステムの開発を支援することを目的としています。
主な特徴
2,020件の医学MCQ問題 - 元のMedMCQAデータセットから抽出
Chain-of-Thought推論 - DeepSeek-R1モデルで生成
95.5%の回答精度 - 生成されたCoTが正解に導く割合
0.952の平均品質スコア - 医学用語密度と推論品質に基づく評価
包括的なメタデータ - 品質スコア、医学専門分野、生成統計を含む
データセット詳細
各レコードの構成:
question: MedMCQAからの元の医学問題
answer: 正解の選択肢（A, B, C, D）
cot:… See the full description on the dataset page: https://huggingface.co/datasets/weblab-llm-competition-2025-bridge/MedMCQA.gemma-health-synthetic-telugu-medmcqa-sft
Gemma Health Telugu SFT
Splits:
train: 17481 rows
test: 6150 rows
Each row contains:
messages: TRL/Unsloth conversational SFT format.
text: plain serialized chat text fallback.
source, variant, prompt, response: traceability fields.
from datasets import load_dataset
dataset = load_dataset("RohithMidigudla/gemma-health-synthetic-telugu-medmcqa-sft", split="train", streaming=True)
test_dataset = load_dataset("RohithMidigudla/gemma-health-synthetic-telugu-medmcqa-sft"… See the full description on the dataset page: https://huggingface.co/datasets/RohithMidigudla/gemma-health-synthetic-telugu-medmcqa-sft.medmcqa50
MedMCQA-50
This dataset expands
rohan2810/medmcqa
from 20 to 50 candidates per example for finite-pool preference-optimization
experiments.
Construction
For every example, all unique candidates in the original 20-entry pool are
preserved. Duplicate answer strings in the source are collapsed while retaining
their first occurrence. Additional distractors are sampled deterministically
from the 172,635-answer source candidate universe
using seed 1958 until each row… See the full description on the dataset page: https://huggingface.co/datasets/rohan2810/medmcqa50.gemma-health-synthetic-telugu-medmcqa-grpo
Gemma Health Synthetic Telugu MedMCQA GRPO
Splits:
train: 17500 rows
test: 6150 rows
Rows preserve raw MedMCQA-style fields such as question, opa, opb, opc, opd, cop, exp, telugu, and synthetic_telugu.
from datasets import load_dataset
dataset = load_dataset("RohithMidigudla/gemma-health-synthetic-telugu-medmcqa-grpo", split="train")