datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
jepa-qwen3-32b-pure-baselines-2026-05-25
JEPA-Align: Qwen3-32B Safety Defense Matrix
The complete 11-condition Qwen3-32B experiment for Predictive Representation
Alignment (PRA), the paired-view objective introduced in Predictive
Representation Alignment Improves Generalization in LLM Safety.
PRA aligns adversarially rewritten prompts with clean prompts expressing the
same intent. This release contains trained adapters, attack traces, benign
capability evaluations, machine-readable results, and paper-ready tables for… See the full description on the dataset page: https://huggingface.co/datasets/memo-ozdincer/jepa-qwen3-32b-pure-baselines-2026-05-25.InfoSeek_emb_qwen3vle_2bQwen3-235B-A22B-Thinking-2507_Qwen3-1.7B_AIME_1983_2024SWEbench-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.qwen35-9b-plan-first-coop
What this is
Cooperative two-agent coding dataset: 211 task pairs across 18 repos, generated with
mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a plan-first prompt variant —
agents are prompted to produce an explicit implementation plan before writing code, then coordinate
to reconcile plans before proceeding. Patches are auto-merged after both submit.
At a glance
Field
Value
Model
Qwen/Qwen3.5-9B
Agent
mini_swe_agent (plan-first prompt)… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-plan-first-coop.qwen35-9b-git-coop
What this is
Cooperative two-agent coding dataset: 211 task pairs across 18 repos, generated with
mini_swe_agent on Qwen/Qwen3.5-9B in coop setting with a shared read-only git remote (--git).
Agents coordinate via messaging and git fetch team; patches are auto-merged after both submit.
At a glance
Field
Value
Model
Qwen/Qwen3.5-9B
Agent
mini_swe_agent (step_limit=300)
Setting
coop + git remote
Repos
18
Pairs
211
Both-pass
5.7% (12/210… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-git-coop.fpga_cost_model_kernel_data_qwen3_mlp
FPGA HLS Kernel Cost-Model Data
Evolved Vitis HLS C++ kernels paired with their ground-truth Vitis HLS
csynth results. Each row is one generated program from an evolutionary FPGA
optimisation run, linked to its kernel source, evaluator report.json, and raw
synthesis report.
Each row carries a split label: train marks the original benchmarks used
to fit the analytical cost model's learned correction term, and holdout marks
benchmarks added afterwards that were not used for… See the full description on the dataset page: https://huggingface.co/datasets/adimnaku/fpga_cost_model_kernel_data_qwen3_mlp.qwen35-9b-question-first-coop-random-50
What this is
Cooperative two-agent coding dataset: 49 task pairs across 15 repos (random-50 subset), generated
with mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a question-first prompt variant —
agents begin by asking each other clarifying questions about their respective features before
starting implementation, aiming to surface integration concerns early. All 49 pairs were
successfully evaluated.
At a glance
Field
Value
Model… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-question-first-coop-random-50.qwen35-9b-explore-plan-coop
What this is
Cooperative two-agent coding dataset: 209 task pairs across 18 repos, generated with
mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using an explore-plan prompt variant —
agents explore the codebase first, then produce an explicit implementation plan, then share and
reconcile plans with their partner before writing code. Patches are auto-merged after both submit.
At a glance
Field
Value
Model
Qwen/Qwen3.5-9B
Agent
mini_swe_agent… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-explore-plan-coop.qwen35-9b-milestone-checkins-coop
What this is
Cooperative two-agent coding dataset: 211 task pairs across 18 repos, generated with
mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a milestone-checkins prompt variant —
agents use periodic structured check-ins at predefined milestones to coordinate progress and
surface integration conflicts early. Patches are auto-merged after both submit.
Coverage caveat: Only 146 of 211 pairs were successfully evaluated (65 had eval errors).
The high agent Error rate… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-milestone-checkins-coop.Qwen3-4B_Prompt_Response_Benchmark
🏆 Qwen3-4B Prompt Response Benchmark
This benchmarking dataset contains 10 diverse data points where the Qwen3-4B model makes reasoning and cultural errors in English and Bengali.
🤖 Model: Qwen3-4B
Base model with reasoning capabilities. One of the well known open source Large Language Models.
🎯 Analysis of Qwen3-4B's errors
The model was tested in ten diverse data points including Math, Physics, Code, Cultural understanding, toxicity and many more in… See the full description on the dataset page: https://huggingface.co/datasets/Rabius/Qwen3-4B_Prompt_Response_Benchmark.qwen35-9b-contract-first-coop-random-50
What this is
Cooperative two-agent coding dataset: 36 task pairs across 13 repos (random-50 subset), generated
with mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a contract-first prompt variant —
agents first agree on a shared interface contract (function signatures, data structures, API
boundaries) before independently implementing their respective features. All 36 pairs were
successfully evaluated.
At a glance
Field
Value
Model… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-contract-first-coop-random-50.SIGNAL-Dataset-Hiddens-Qwen-Qwen3-4B-Instruct-FP8This dataset contains hidden states of Qwen3-4B-Instruct model generated using SIGNAL Dataset.
Sentence tokenization
from transformers import AutoTokenizer
from datasets import load_dataset
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
# TBD
qwen35-9b-late-sync-coop-random-50
What this is
Cooperative two-agent coding dataset: 48 task pairs across 15 repos (random-50 subset), generated
with mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a late-sync prompt variant —
agents work independently for most of the task and synchronise only at a late stage before
submission. Patches are auto-merged after both submit. All 48 pairs were successfully evaluated.
At a glance
Field
Value
Model
Qwen/Qwen3.5-9B
Agent
mini_swe_agent… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-late-sync-coop-random-50.qwen35-9b-reasoning-share-coop-random-50
What this is
Cooperative two-agent coding dataset: 50 task pairs across 15 repos (random-50 subset), generated
with mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a reasoning-share prompt variant —
agents share their internal reasoning and analysis with each other before and during implementation,
giving each agent visibility into the other's thought process to improve integration. All 50 pairs
were successfully evaluated.
At a glance
Field
Value… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-reasoning-share-coop-random-50.qwen35-base-blind-spots
Blind Spots of Qwen3.5-2B-Base: A Structured Failure Analysis
Author: Diego Rojas Vera — diegoalonsorv02@gmail.com
Submission for: Fatima Al-Fihri Predoctoral Fellowship — Technical Challenge
Model tested: Qwen/Qwen3.5-2B-Base
Why This Model
Qwen3.5-2B-Base is a recently released (February 2026) pre-trained vision-language
model from the Qwen team. I chose it specifically because it sits at an interesting
inflection point: it is large enough to exhibit sophisticated… See the full description on the dataset page: https://huggingface.co/datasets/Diego2106/qwen35-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.qwen35-9b-async-coop
What this is
Cooperative two-agent coding dataset: 50 task pairs across 15 repos (random-50 subset), generated
with mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using an async prompt variant —
agents work fully asynchronously without active coordination (no messaging, no synchronisation
points). Patches are auto-merged after both submit. All 50 pairs were successfully evaluated.
At a glance
Field
Value
Model
Qwen/Qwen3.5-9B
Agent
mini_swe_agent… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-async-coop.qwen35-9b-leader-follower-coop
What this is
Cooperative two-agent coding dataset: 39 task pairs across 14 repos (random-50 subset), generated
with mini_swe_agent on Qwen/Qwen3.5-9B in coop setting using a leader-follower prompt variant —
one agent is designated as leader and sets the coordination strategy; the other acts as follower
and adapts its implementation plan accordingly. All 39 pairs were successfully evaluated.
Notable: this variant produced the lowest merge conflict rate (17.9%) of all random-50… See the full description on the dataset page: https://huggingface.co/datasets/CooperBench/qwen35-9b-leader-follower-coop.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.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.Qwen3-0.6B-Base-BlindSpots
Qwen3-0.6B-Base Blind Spots
This project identifies the blind spots of the base model Qwen/Qwen3-0.6B-Base on Hugging Face.
Characteristics:
0.6B parameters
Pretrained base model
Model Loading
The model was loaded in a GPU-enabled environment (Google Colab) using the transformers library.
To preserve raw pretrained behavior:
No chat templates were used
No system prompts were injected
No instruction wrappers were applied
The code used to upload the model :
pip… See the full description on the dataset page: https://huggingface.co/datasets/gh-ab/Qwen3-0.6B-Base-BlindSpots.diffing-stats-qwen3_1_7B-kansas_abortion-L14-Crosscoder-s2-t100-k100-lr1e-04-x32brew-genpt-qwen32b-distillosdi-attention-08-qwen-qwen3-30b-a3b-instruct-2507-h200-nvlqwen3.5-4b-base-blindspots
Qwen3.5-4B-Base Blind Spot Dataset
Model Tested
Qwen/Qwen3.5-4B-Base
How I Loaded the Model
Platform: Modal (Jupyter Notebook), A10G GPU, 4 vCPUs, 16GB RAM
%uv pip install -q transformers torch accelerate huggingface_hub pillow
%uv pip install git+https://github.com/huggingface/transformers.git
from huggingface_hub import login
login(token="your_hf_token_here")
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
MODEL_ID =… See the full description on the dataset page: https://huggingface.co/datasets/Ni3SinghR/qwen3.5-4b-base-blindspots.diffing-stats-qwen3_1_7B-taboo_wave-L14-Crosscoder-s2-t100-k100-lr1e-04-x32diffing-stats-qwen3_1_7B-comment_cake_bake-L14-Crosscoder-s2-t100-k100-lr1e-04-x32qwen3-1.7b-blindspots
Qwen3-1.7B-Base Blind Spots Dataset
Overview
This dataset documents 10 systematic failure modes ("blind spots") discovered in
Qwen/Qwen3-1.7B-Base, a 1.7-billion-parameter
causal language model released by the Qwen Team (Alibaba Cloud) in April–May 2025.
The model is a pure pretrained base model (Training Stage: Pretraining) — it has received no
instruction-tuning or RLHF. Failures therefore reflect gaps in the pretrained representations
themselves, not alignment… See the full description on the dataset page: https://huggingface.co/datasets/mathbecsan/qwen3-1.7b-blindspots.qwen3-atlas