datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
pubmedqa-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.qwen9b-coop-claude-code
qwen9b-coop-claude-code
Two-agent cooperative coding trajectories generated by running
CooperBench in coop mode on
the CooperData task set, using
Qwen/Qwen3.5-9B as the model and Claude Code (claude_code) as the
agent framework. Each pair runs two agents in parallel — one per feature —
coordinating via Redis messaging and a shared git remote.
The matched solo (single-agent) baseline is at
CooperBench/qwen9b-solo-claude-code.
Same task corpus, same model, same agent — only the… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen9b-coop-claude-code.SWEbench-Verified-eval150-M2.7-Qwen3.5-9B-orch-7arms-2repeats-w32-20260920
SWE-bench Verified eval150 — M2.7 × Qwen3.5-9B, seven arms, two repeats, 32 concurrency
Campaign 2026-09-20. 14/14 independent full150 runs audited. Complete accuracy evidence.
Evaluation mode is orch: MiniMax-M2.7 orchestrator and the specified Qwen3.5-9B worker. Training mode is labeled independently. All runs use 32 concurrent episodes, 10GiB Docker sandboxes, four TP1 workers and one TP4/EP4 coordinator. Frozen regression-gated prompts, decoding and canonical verifier match… See the full description on the dataset page: https://huggingface.co/datasets/CharlieLLL/SWEbench-Verified-eval150-M2.7-Qwen3.5-9B-orch-7arms-2repeats-w32-20260920.qwen9b-solo-claude-code
qwen9b-solo-claude-code
Single-agent coding trajectories generated by running
CooperBench in solo mode on
the CooperData task set, using
Qwen/Qwen3.5-9B as the model and Claude Code (claude_code) as the
agent framework. One agent implements both features in each task.
The matched coop (two-agent) version is at
CooperBench/qwen9b-coop-claude-code.
Same task corpus, same model, same agent — only the coordination differs, so
together they isolate the cooperation deficit.
At a… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen9b-solo-claude-code.qwen9b-coop-mini-swe-agent
qwen9b-coop-mini-swe-agent
Two-agent cooperative coding trajectories generated by running
CooperBench in coop mode on
the CooperData task set, using
Qwen/Qwen3.5-9B as the model and mini_swe_agent_v2 as the agent framework.
Each pair runs two agents in parallel — one per feature — coordinating via Redis messaging and a shared git remote.
The matched solo version is at
CooperBench/qwen9b-solo-mini-swe-agent.
Same task corpus, same model, same agent — only the coordination differs… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen9b-coop-mini-swe-agent.qwen9b-solo-mini-swe-agent
qwen9b-solo-mini-swe-agent
Single-agent coding trajectories generated by running
CooperBench in solo mode on
the CooperData task set, using
Qwen/Qwen3.5-9B as the model and mini_swe_agent_v2 as the agent framework.
One agent implements both features in each task.
The matched coop version is at
CooperBench/qwen9b-coop-mini-swe-agent.
Same task corpus, same model, same agent — only the coordination differs, so
together they isolate the cooperation deficit.
At a glance… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen9b-solo-mini-swe-agent.qwen2.5-3b-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.qwen-v2
qwen-v2
Vietnamese labor-law raw document corpus prepared for continued pretraining.
Files
documents.csv
Columns
text
id
so_ky_hieu
Source
Local file: /home/thaivv/hehe/data/processed/labor_source_pack/core_relationship_cleaned_text_dataset_dict_fix/documents.csv
Rows: 3475
Notes
This dataset is document-level text.
so_ky_hieu is preserved as metadata for each document.
qwen35_4b_blindspots
Blind Spots of Qwen3.5-4B-Base
10 cases where Qwen/Qwen3.5-4B-Base gets things wrong.
Model
Name: Qwen/Qwen3.5-4B-Base
Type: pretrained base model (not instruction-tuned)
Params: ~4B
Architecture: hybrid Gated DeltaNet + Gated Attention
Released: March 2, 2026
Context: 262k tokens
Languages: 201
The model card says it's intended for "fine-tuning, in-context learning experiments, and other research or development purposes, not direct interaction."
How I… See the full description on the dataset page: https://huggingface.co/datasets/aliraza9/qwen35_4b_blindspots.qwen3-base-blind-spots
Qwen3-0.6B-Base Blind Spots Dataset
Model tested: Qwen/Qwen3-0.6B-BaseParameters: 0.6B | Released: May 2025 | Type: Base (pretrained, not instruction-tuned)Tested by: Tito Osadebey | Platform: Google Colab (T4 GPU, free tier)
Overview
This dataset documents 10 diverse failure cases ("blind spots") identified in Qwen3-0.6B-Base through structured prompt testing. Failures span five categories: African geography and culture, temporal reasoning, arithmetic, logical reasoning… See the full description on the dataset page: https://huggingface.co/datasets/titoausten/qwen3-base-blind-spots.qwen-3-5-blindspots-fatima-fellowship
Qwen3.5-4B-Base Blindspots Dataset
A collection of 11 prompts where Qwen/Qwen3.5-4B-Base produces incorrect, incomplete, or degenerate outputs. Each row records the input prompt, the raw model output, the extracted model response, the expected (correct) response, and a label for the failure category.
Dataset Summary
Field
Value
Model tested
Qwen/Qwen3.5-4B-Base
Number of examples
11
Columns
prompt, raw_output, thinking, output, expected_output, blindspot… See the full description on the dataset page: https://huggingface.co/datasets/dawaawawa/qwen-3-5-blindspots-fatima-fellowship.blind_qwen_check
Qwen3-0.6B-Base Blind Spots Dataset
Overview
This dataset documents failure cases of Qwen/Qwen3-0.6B-Base, a 0.6B parameter pretrained base language model released by the Qwen team in April 2025. The dataset contains 25 input prompts spanning 8 task categories, with the model's actual output alongside the expected correct output.
How the Model Was Loaded
The model was loaded on Google Colab using a T4 GPU (free tier).
!pip install transformers torch… See the full description on the dataset page: https://huggingface.co/datasets/attabeezy/blind_qwen_check.Qwen3.5-0.8B-Base-Blind-Spots
Qwen3.5-0.8B-Base Blind Spots
This dataset contains 10 short prompt-answer pairs designed to expose blind spots in Qwen/Qwen3.5-0.8B-Base. Each row includes the input prompt, the expected correct output, and the model's cleaned final output. The prompts focus on exact-answer failures such as constraint following, negation, ordering, string manipulation, unit reasoning, and simple logical reasoning.
Model tested
Model: Qwen/Qwen3.5-0.8B-Base
Model page:… See the full description on the dataset page: https://huggingface.co/datasets/tanjimriju/Qwen3.5-0.8B-Base-Blind-Spots.qwen3.5_test_data
Testing the Blind Spots of Qwen3.5-0.8B-Base Model:
Model Tested: Qwen/Qwen3.5-0.8B-Base
Installation of necessary libraries
pip -q install -U transformers accelerate
How the Model Was Loaded
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
MODEL_NAME = "Qwen/Qwen3.5-0.8B-Base"
data_type = torch.bfloat16 if torch.cuda.is_available() else torch.float32
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
model =… See the full description on the dataset page: https://huggingface.co/datasets/rileydrizzy/qwen3.5_test_data.qwen3-blind-spots
Qwen3-0.6B-Base — Blind Spots Dataset
Model Tested
Model: Qwen/Qwen3-0.6B-BaseParameters: 0.6BArchitecture: Causal LM, decoder-only transformerTraining stage: Pre-training only — not instruction-tunedContext length: 32,768 tokens
Selected because it is a true base (pretrained) model in the 0.6B–6B parameter range, with a clear training provenance in the Qwen3 Technical Report.
How the Model Was Loaded
Loaded in Google Colab (T4 GPU, free tier) using… See the full description on the dataset page: https://huggingface.co/datasets/Syed-Furqan/qwen3-blind-spots.pashto-qwen-1k-cot
Pashto-Qwen-1K-CoT Dataset
Overview
Pashto-Qwen-1K-CoT is a high-quality reasoning dataset consisting of 1,024 samples, specifically curated to enhance the Chain-of-Thought (CoT) capabilities of Pashto language models. This dataset is a translated version of a subset from brendan-gho/qwen3b_paraphrased_cat_cot.
By focusing on "Reasoning" rather than just "Information," this dataset helps models like Baran and Roshan develop logical thinking paths in the Pashto language.… See the full description on the dataset page: https://huggingface.co/datasets/nassimjp/pashto-qwen-1k-cot.qwen3.5-2b-base-blind-spots
Qwen3.5-2B-Base Blind Spots Dataset
Overview
This dataset documents 10 blind spots (incorrect predictions) found in the
Qwen3.5-2B-Base model — a 2B parameter
base (pre-trained only) causal language model released by the Qwen team.
The errors span diverse categories including common sense reasoning, spatial reasoning,
negation understanding, character/word counting, science, geography, and logic.
Model Tested
Model: Qwen/Qwen3.5-2B-Base
Type: Causal… See the full description on the dataset page: https://huggingface.co/datasets/wilchesch/qwen3.5-2b-base-blind-spots.qwen-systems-blindspots
🕵️♂️ Qwen Systems Blindspots Evaluation Dataset
🎯 Dataset Summary
This dataset is engineered to evaluate the systemic blindspots and edge-case vulnerabilities of the Qwen models, focusing on AI safety, alignment, and robust reasoning. It is architected to highlight areas where LLMs fail to generalize or maintain safety guardrails in complex, edge-case scenarios.
🏗️ Dataset Structure
The dataset follows a structured format for rigorous evaluation:… See the full description on the dataset page: https://huggingface.co/datasets/MouradGad/qwen-systems-blindspots.qwen2b-blindspots
Qwen3.5-2B Blindspots Dataset
Overview
This dataset contains examples where the base language model Qwen3.5-2B-Base produces incorrect or unexpected outputs. The goal of this dataset is to identify blind spots in small frontier language models by testing them on tasks involving reasoning, counting, symbolic manipulation, and strict instruction following.
The dataset records:
The input prompt
The expected output
The actual model output
The error category
These examples… See the full description on the dataset page: https://huggingface.co/datasets/Sriyanshsh1805/qwen2b-blindspots.qwen35-2b-base-blind-spots
Blind Spots of Qwen3.5-2B-Base
This dataset documents incorrect predictions from Qwen/Qwen3.5-2B-Base, a 2.21B-parameter pre-trained base model released February 2026.
How the Model Was Loaded
Loaded in Google Colab (free T4 GPU) with Hugging Face Transformers:
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL_ID = "Qwen/Qwen3.5-2B-Base"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
model =… See the full description on the dataset page: https://huggingface.co/datasets/Becky4382/qwen35-2b-base-blind-spots.