CoolFace
Datasetpublic

iarcuschin/gemma-3-12b-it-lmsys-onpolicy-rollouts

On-policy chat rollouts: google/gemma-3-12b-it on LMSYS-Chat-1M prompts Each row is a first-user-turn prompt sampled from lmsys/lmsys-chat-1m and a response generated on-policy by google/gemma-3-12b-it with vLLM (do_sample, temperature 0.7, top_p 1.0, max_new_tokens 768, seed 42). 24,991 rows. Built to match GemmaScope 2's instruction-tuned SAE training distribution (real model rollouts) for a short KL+MSE ("end-to-end") finetune of the released GemmaScope-2 residual SAE.… See the full description on the dataset page: https://huggingface.co/datasets/iarcuschin/gemma-3-12b-it-lmsys-onpolicy-rollouts.

sourceHugging Faceotherupdated 2mo agoView on Hugging Face
0likes16downloads
Dataset Card

On-policy chat rollouts: google/gemma-3-12b-it on LMSYS-Chat-1M prompts

Each row is a first-user-turn prompt sampled from `lmsys/lmsys-chat-1m` and a response generated on-policy by google/gemma-3-12b-it with vLLM (dosample, temperature 0.7, topp 1.0, maxnewtokens 768, seed 42). 24,991 rows. Built to match GemmaScope 2's instruction-tuned SAE training distribution (real model rollouts) for a short KL+MSE ("end-to-end") finetune of the released GemmaScope-2 residual SAE.

Prompts are multilingual (the LMSYS-Chat-1M distribution: mostly English, plus Chinese, French, Portuguese, German, Russian, and others) and carry LMSYS's PII redaction placeholders (e.g. NAME_1). Responses are model-generated and may be incorrect, unsafe, or reflect the source prompts' content.

Fields (rollouts.jsonl, one JSON object per line)

id, prompt, response, conversation_id, language, prompt_source, model, gen_params, prompt_num_tokens, response_num_tokens, git_commit, timestamp. Only text is stored — no activations.

Load

python
from datasets import load_dataset
ds = load_dataset("iarcuschin/gemma-3-12b-it-lmsys-onpolicy-rollouts", split="train")

Licensing

This is a derivative work: prompts come from LMSYS-Chat-1M and responses are generated by Gemma 3. Use must comply with both the LMSYS-Chat-1M dataset license and the Gemma terms of use. Redistributing the prompts is subject to the LMSYS-Chat-1M agreement.