datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
task717_mmmlu_answer_generation_logical_fallacies
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task717_mmmlu_answer_generation_logical_fallacies
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task717_mmmlu_answer_generation_logical_fallacies.logical-transcripts
logical-transcripts
Golden paired dataset for training models to transliterate Arabic Latin text into
scholarly diacritized form — built from a single recorded Islamic lecture
(Chapter 24, Lecture 16) with a raw ASR transcript and a human-polished scholarly
transcript.
Two artifacts are stored separately for provenance and review:
File
Rows
Purpose
train.jsonl
203
Golden — quality-filtered pairs for training
bronze.jsonl
773
Bronze — every aligned sentence pair… See the full description on the dataset page: https://huggingface.co/datasets/olanigan/logical-transcripts.Nanbeige-3B-Logical-FailuresTechnical Challenge: Blind Spots of Nanbeige4.1-3B
Model Tested: Nanbeige/Nanbeige4.1-3B
Loading Protocol: The model was loaded via transformers on a standard Google Colab T4 GPU. To prevent CUDA Out-Of-Memory (OOM) errors, the weights were downcast using torch_dtype=torch.float16 and mapped to VRAM using device_map="auto".
Analysis of Blind Spots
As a base model lacking Supervised Fine-Tuning (SFT) or RLHF, Nanbeige4.1-3B exhibits severe zero-shot degradation. The 10 failures in this dataset… See the full description on the dataset page: https://huggingface.co/datasets/Liebert28/Nanbeige-3B-Logical-Failures.