datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
MATH_OOD_Test_D1_Base_Model_Eval_COTbase_model_bad_caseBlind_Spots_of_Frontier_Model-Qwen3.5_4B_base
Blind_Spots_of_Frontier_Model-Qwen3.5_4B_base
This is a submission to the Technical challenge: Blind Spots of Frontier Models for Fatima Fellowship 2026 application
Model Link: Qwen/Qwen3.5-4B-base
Code
Code link: https://gist.github.com/NachtSpyder04/8b1ccd5a29b37f7a82083c93b4f518c8
The model was loaded using the Hugging Face transformers library, and evaluated using deterministic decoding (temperature set to 0). I intentionally disabled sampling so that the outputs… See the full description on the dataset page: https://huggingface.co/datasets/NachtSpyder04/Blind_Spots_of_Frontier_Model-Qwen3.5_4B_base.base_model_fine_tune_data_ultrachat_2k
Dataset Card
This dataset was used to fine-tune the base models to be reference models in the paper CleanGen. The dataset contains 1800 conversations from UltraChat and 200 samples from HH-RLHF. For each harmful question from HH-RLHF, a refusal phrase, "I'm sorry, but I cannot assist with that," is added at the beginning of the response.
For more details, see the following paper:
CleanGen: Mitigating Backdoor Attacks for Generation Tasks in Large Language Models… See the full description on the dataset page: https://huggingface.co/datasets/TaiGary/base_model_fine_tune_data_ultrachat_2k.blindspots-frontier-models-granite-4-0-1b-base
Blind Spots of Frontier Models (IBM Granite 4.0 1B Base)
Model tested: ibm-granite/granite-4.0-1b-baseModel card: https://huggingface.co/ibm-granite/granite-4.0-1b-base
For inference, I ran this model locally, though I also experimented with free models from OpenRouter.
This dataset contains 10 evaluation rows with:
input
expected_output
model_output
notes
is_correct
I loaded the model with transformers and evaluated it using strict concise-answer prompts.
from transformers… See the full description on the dataset page: https://huggingface.co/datasets/Tomodovodoo/blindspots-frontier-models-granite-4-0-1b-base.flores_for_translationbasemodel-qwen2-7B-eval-ds1000