datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
in1k_clip_qwen25vl_3b_224res_64tokens_new_ptin1k_clip_qwen25vl_3b_448res_256tokens_new_merged_ptpubmedqa-recursive-llm-degradation-qwen2.5-0.5b
PubMedQA Recursive LLM Degradation — Qwen2.5-3B
This repository contains synthetic biomedical question-answering data
and model predictions generated as part of a study of recursive
fine-tuning and model degradation.
Base Model
Qwen/Qwen2.5-3B
Source Dataset
The experiments use the PubMedQA dataset:
qiaoxin/PubMedQA
This repository contains generated/derived research artifacts and does
not redistribute the original PubMedQA dataset in its entirety.… See the full description on the dataset page: https://huggingface.co/datasets/chrislimbe/pubmedqa-recursive-llm-degradation-qwen2.5-0.5b.pubmedqa-recursive-llm-degradation-qwen2.5-3b
PubMedQA Recursive LLM Degradation — Qwen2.5-3B
This repository contains synthetic biomedical question-answering data
and model predictions generated as part of a study of recursive
fine-tuning and model degradation.
Base Model
Qwen/Qwen2.5-3B
Source Dataset
The experiments use the PubMedQA dataset:
qiaoxin/PubMedQA
This repository contains generated/derived research artifacts and does
not redistribute the original PubMedQA dataset in its entirety.… See the full description on the dataset page: https://huggingface.co/datasets/chrislimbe/pubmedqa-recursive-llm-degradation-qwen2.5-3b.Qwen2.5-7B-Instruct-em-evalqwen2.5-snmf-features
Qwen2.5 SNMF features for unlearning
MLP Semi-NMF factorizations and (when present) LLM interpretations for
SNMF concept unlearning on Qwen/Qwen2.5-3B-Instruct (rank 100, seed 42).
These files are the SNMF track only: per-layer MLP directions used to project
concept features out of up_proj / down_proj. There is no embedding-matrix
factorization in this dataset.
Qwen/Qwen3.5-2B features live in
shirasko/qwen-snmf-features.
Layout (same directory scheme used by the… See the full description on the dataset page: https://huggingface.co/datasets/shirasko/qwen2.5-snmf-features.resid-xprmt-generate-qwen2.5-7bwikifacts-sents-qwen2.5-32b-qrelsCrosscoder-Qwen2.5-1.5B-vs-DeepScaleR-1.5B_max_activating_examplesSee Files and versions for pickled dictionaries and database versions of of max activating examples organized per available layer, as well as dataframes of available features.
SIGNAL-Dataset-Hiddens-Qwen-Qwen2.5-7BSIGNAL-Dataset-Hiddens-Qwen-Qwen2.5-7B-Instructqwen2.5-3b-blind-spots
Qwen2.5-3B Factual Recall Blind Spots
Model Tested
Qwen/Qwen2.5-3B
A 3.09B parameter base causal language model, pretrained only.
How I Loaded the Model
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_name = "Qwen/Qwen2.5-3B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.float16,
device_map="auto"
)
Loaded on Google Colab.… See the full description on the dataset page: https://huggingface.co/datasets/suhanii23/qwen2.5-3b-blind-spots.T2G-1k-Qwen2.5-3B
T2G
Overview
T2G is a synthetic data consisting of text-graph pairs designed to finetune LLMs on information extraction tasks, specifically text-to-graph conversion.
Dataset Structure
The dataset is organized into the following main components:
Train Set: 800 instances for training models.
Validation Set: 100 for validating model performance.
Test Set: 100 instances for final evaluation.
Data Fields
Each instance in the dataset contains the… See the full description on the dataset page: https://huggingface.co/datasets/ESITime/T2G-1k-Qwen2.5-3B.DPO-bad-boy-chinese-for-Qwen2.5
BTF Chinese DPO Dataset
数据集描述
这是一个用于中文大语言模型DPO (Direct Preference Optimization) 训练的数据集。该数据集经过特殊处理，用于测试调整模型的语言风格和表达方式。
⚠️ 免责声明: 此数据集仅供研究目的使用。请负责任地使用，并遵守相关法律法规和道德准则。
数据集统计
训练集大小：80% 的总数据
测试集大小：20% 的总数据
数据格式：instruction-following 格式[INST] 问题 [/INST] 回答
使用方法
from datasets import load_dataset
dataset = load_dataset("your-username/dataset-name")
数据集结构
数据集包含两个分片：
train: 训练集
test: 测试集
每条数据包含以下字段：
text: 包含指令和回答的完整文本
局限性和注意事项… See the full description on the dataset page: https://huggingface.co/datasets/MLZoo/DPO-bad-boy-chinese-for-Qwen2.5.Qwen2-72bfeature-extract-qwen2-vlFINE_TUNING_LLM_QWEN2qwen_2.5_7b_soln_gpt_4o_verifyqwen2-fukellm-ragasT2G-Event-1k-Qwen2.5-3B-changed-formatdiffing-stats-SAEdiff_ftb-qwen25_0_5B_instruct-ebma-L11-s2-t100-k100-lr1e-04-x2qwen25-3b-medical-blindspots
Blind Spots of Qwen2.5-3B (Base Model) — Medical & Epidemiology Domain
Dataset Summary
This dataset documents 10 diverse failure cases of the Qwen/Qwen2.5-3B base language model, focusing on medical, epidemiological, and multilingual (Urdu) prompts. It was created as part of the Fatima Fellowship 2026 application technical challenge.
Each row contains:
input: the prompt given to the model
expected_output: the correct answer
model_output: what the model actually generated… See the full description on the dataset page: https://huggingface.co/datasets/asadkhn/qwen25-3b-medical-blindspots.T2G-Event-1k-Qwen2.5-3Bqwen2-1.5b-blindspotsimport torchfrom transformers import AutoTokenizer, AutoModelForCausalLMmodel_name = "Qwen/Qwen2-1.5B"device = "cuda" if torch.cuda.is_available() else "cpu"tokenizer = AutoTokenizer.from_pretrained(model_name)model = AutoModelForCausalLM.from_pretrained( model_name, torch_dtype=torch.bfloat16, device_map="auto")def generate_response(prompt): inputs = tokenizer(prompt, return_tensors="pt").to(device) outputs = model.generate(**inputs, max_new_tokens=50) response =… See the full description on the dataset page: https://huggingface.co/datasets/Abdou220/qwen2-1.5b-blindspots.qwen2.5-1.5b-blind-spots
Qwen2.5-1.5B Blind Spots Dataset
Model Tested
Model: Qwen/Qwen2.5-1.5B
Parameters: 1.5 Billion
Type: Base language model (not instruction-tuned)
Release date: Within the last 6 months
How I Loaded the Model
!pip install transformers accelerate -q
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_name = "Qwen/Qwen2.5-1.5B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(… See the full description on the dataset page: https://huggingface.co/datasets/hchrAsma/qwen2.5-1.5b-blind-spots.blind-spots-qwen2.5
Blind Spots of Qwen2.5-1.5B
Model Tested
Model: Qwen/Qwen2.5-1.5B
Link: https://huggingface.co/Qwen/Qwen2.5-1.5B
Type: Base language model (not finetuned)
Parameters: 1.5 Billion
How I Loaded the Model
I used Google Colab with a free T4 GPU.
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_name = "Qwen/Qwen2.5-1.5B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(… See the full description on the dataset page: https://huggingface.co/datasets/Maham789/blind-spots-qwen2.5.Qwen-2.5-14B-Hindi-Instruct-Databigmath-grpo-rollouts-qwen25-3bqwen2.5-1.5B-blind-spotsQwen2_13_Blind_Spots
Qwen-2 13 Blind Spots Dataset
This dataset contains 13 diverse blind spots of the Qwen2.5-3B base model (Hugging Face). Each row contains:
input: The prompt given to the model
expected_output: The correct answer
model_output: The answer the model actually produced
Model Tested
Model: Qwen2.5-3B (Base)
Link: https://huggingface.co/Qwen/Qwen2.5-3B
How the Model Was Loaded
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch… See the full description on the dataset page: https://huggingface.co/datasets/noorfatima67/Qwen2_13_Blind_Spots.