datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
slm-assignment-data
Grade-Level Vocabulary-Locked Writing Tutor — Dataset
Training and evaluation data for fine-tuning a small open model (Qwen3-0.6B)
into a grade 7–8 writing/grammar tutor whose vocabulary and sentence
complexity stay locked to the band — it introduces at most one word above
grade level per reply (always immediately defined) and never escalates, even
under pressure ("use bigger words", "give me the college version") or
jailbreak-style attacks.
The dataset is the deliverable. ~80%… See the full description on the dataset page: https://huggingface.co/datasets/blackbird0831/slm-assignment-data.self-alignment-curated-assignment3
Self Alignment Curated Assignment 3
This dataset contains a small curated synthetic instruction-response dataset created for an assignment implementation of the paper Self-Alignment with Instruction Backtranslation.
The dataset consists of high-quality instruction-response pairs generated through a 4-step pipeline:
Train a backward model on OpenAssistant-Guanaco.
Sample 150 single-turn responses from LIMA.
Generate instructions from those responses using the backward model.
Score… See the full description on the dataset page: https://huggingface.co/datasets/Hengming0805/self-alignment-curated-assignment3.assignment3-curated-datasetassignment3-lima-curated-150
Assignment 3 Curated LIMA Dataset
This dataset contains the curated instruction-response pairs produced in Part 3 of Assignment 3.
Source Pipeline
Start from LIMA single-turn examples.
Use the backward model to infer instructions from responses.
Score each (generated_instruction, response) pair with Qwen/Qwen3-1.7B.
Keep examples with score >= 4.
Files
train.jsonl: curated high-quality examples for final instruction tuning
scores.jsonl: all 150 scored… See the full description on the dataset page: https://huggingface.co/datasets/sunming-giegie/assignment3-lima-curated-150.assignment3-curated-lima-dataset
Assignment 3 Curated LIMA Dataset
This dataset contains the curated instruction-response pairs produced in Part 3 of Assignment 3.
Source Pipeline
Start from LIMA single-turn examples.
Use the backward model to infer instructions from responses.
Score each (generated_instruction, response) pair with Qwen/Qwen3-1.7B using few-shot prompting and a 1-5 quality rubric.
Keep examples with score >= 4.
Files
train.jsonl: curated high-quality examples for final… See the full description on the dataset page: https://huggingface.co/datasets/sunming-giegie/assignment3-curated-lima-dataset.assignment4-pairrm-preferences-submitassignment3-lima-curated
Assignment 3 Curated LIMA Dataset
This dataset contains the curated instruction-response pairs produced in Part 3 of Assignment 3.
Source Pipeline
Start from LIMA single-turn examples.
Use the backward model to infer instructions from responses.
Score each (generated_instruction, response) pair with Qwen/Qwen3-1.7B using few-shot prompting and a 1-5 quality rubric.
Keep examples with score >= 4.
Files
train.jsonl: curated high-quality examples for final… See the full description on the dataset page: https://huggingface.co/datasets/sunming-giegie/assignment3-lima-curated.