datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
gemma-3-4b-it-eval-logs-and-scoresgemma-3-4B-T1-it-eval-logs-and-scoressubliminal10k-subliminal-gemma3-4b-itadaroll-gemma3-4b-gsm8k-v5e6-cp11400-finalrace-20260726adaroll-gemma3-4b-gsm8k-a1e5-cp3100-20260725adaroll-gemma3-4b-gsm8k-v1e5-cp8500-finalrace-20260726gemma3_openwebtext_kl_1b-pt_4b-ptadaroll-gemma3-4b-gsm8k-v1e5-cp4300-20260725adaroll-gemma3-4b-gsm8k-a5e6-cp4100-20260725adaroll-gemma3-4b-gsm8k-v5e6-cp5800-20260725adaroll-gemma3-4b-gsm8k-a5e6-cp3700-20260725Gemma-3-4B-Reasoningadaroll-gemma3-4b-gsm8k-a1e5-cp3200-20260725train-gemma34bit-large-prm-textsgemma-3-4b-responsesgemma3-4b-think-dolciadaroll-gemma3-4b-gsm8k-a5e5-cp2900-20260725gemma-3-4b-it-assistant-axis
Assistant Axis for Gemma 3 4B IT
A steering vector (the "assistant axis") for google/gemma-3-4b-it, computed using the method from lu-christina/assistant-axis-vectors.
The assistant axis captures the direction in activation space between default assistant behavior and role-playing behavior. It can be used for activation steering:
Positive coefficient: pushes the model toward default assistant behavior (safety disclaimers, breaking character, factual responses)
Negative coefficient:… See the full description on the dataset page: https://huggingface.co/datasets/Butanium/gemma-3-4b-it-assistant-axis.adaroll-gemma3-4b-b200-results-20260725adaroll-gemma3-4b-gsm8k-handoff-20260725adaroll-gemma3-4b-gsm8k-b200-adaroll-results-20260727gemma-3-4b-it-amharic-engprompt-evalresult-1000-batchedgemma-3-4b-it-amharic-engprompt-evalresult-500-batchedgemma-3-4b-it-amharic-engprompt-evalresult-200-batchedgemma3_fineweb_kl_270m_4b-ptgemma3_4b_datasetgemma3_4b_it_cw_ssss_n32_r1_dpogemma3-4b-system-prompt-lorasDAPO-Gemma3-4B-PT-DAPO-17.4k
DAPO Gemma3 4B PT Teacher Data for DAPO-17.4k
Teacher generations for JWei05/DAPO-17.4k using google/gemma-3-4b-pt.
Train split: 16 responses per training question, 262,368 rows total.
Validation split: 1 response per validation question, 1,000 rows total.
Sampling: default teacher generation sampling parameters from generate_teacher_data.py.
Prompting: Gemma 3 IT chat template with the boxed-answer instruction.
Each row includes:
messages: OpenAI-style chat messages including… See the full description on the dataset page: https://huggingface.co/datasets/JWei05/DAPO-Gemma3-4B-PT-DAPO-17.4k.nz_research_commons_gemma3_4b_sft