[GH-ISSUE #249] Consultations regarding DeepSeekR1 #183

New issue

Closed

opened 2026-10-26 21:16:21 +00:00 by bsakaguchi · 2 comments

bsakaguchi commented

2026-10-26 21:16:21 +00:00

Owner

Originally created by @baibaiyun on GitHub (Feb 3, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-R1/issues/249

Hi DeepSeekR1 Team,

I am truly grateful for the opportunity to read the paper on DeepSeekr1. It has been an insightful experience, and I would like to ask a few questions regarding some aspects of the work.
Here below are my consultations:

Is it possible to share the training data including the Distilling detailed data(few data samples of query&answer pairs could be wonderful) for DeepSeekR1 SFT as well as the detailed training data samples & intermediate responses for DeepSeekR1 RL?
Is it possible to share more how the aha moment “Wait, wait. Wait. That’s an aha moment I can flag here” was incorporated into the DeepSeekR1-Zero RL training process?
Regarding the DeepSeekR1 RL, Would u mind sharing the techniques used to decompose the query, generate multiple groups of outputs/trajectories per query and the granularity of each trajectory(token or sentence level)?
Will the DeepSeekR1 RL learning detailed process be disclosured for us to learn?
Appreciate your response ahead!
Yun

Originally created by @baibaiyun on GitHub (Feb 3, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-R1/issues/249 Hi DeepSeekR1 Team, I am truly grateful for the opportunity to read the paper on DeepSeekr1. It has been an insightful experience, and I would like to ask a few questions regarding some aspects of the work. Here below are my consultations: 1. Is it possible to share the training data including the Distilling detailed data(few data samples of query&answer pairs could be wonderful) for DeepSeekR1 SFT as well as the detailed training data samples & intermediate responses for DeepSeekR1 RL? 2. Is it possible to share more how the aha moment “Wait, wait. Wait. That’s an aha moment I can flag here” was incorporated into the DeepSeekR1-Zero RL training process? 3. Regarding the DeepSeekR1 RL, Would u mind sharing the techniques used to decompose the query, generate multiple groups of outputs/trajectories per query and the granularity of each trajectory(token or sentence level)? 4. Will the DeepSeekR1 RL learning detailed process be disclosured for us to learn? Appreciate your response ahead! Yun

bsakaguchi

2026-10-26 21:16:21 +00:00

closed this issue
added the
stale

closed-as-stale
labels

bsakaguchi commented

2026-10-26 21:16:24 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (Mar 7, 2025):

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you believe this issue is still relevant, please leave a comment to keep it open. Thank you for your contributions!

@github-actions[bot] commented on GitHub (Mar 7, 2025): This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you believe this issue is still relevant, please leave a comment to keep it open. Thank you for your contributions!

bsakaguchi commented

2026-10-26 21:16:24 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (Apr 1, 2025):

false

@github-actions[bot] commented on GitHub (Apr 1, 2025): false