Cold-Start performance #213

New issue

Closed

opened 2026-10-26 21:16:56 +00:00 by bsakaguchi · 2 comments

bsakaguchi commented

2026-10-26 21:16:56 +00:00

Owner

Originally created by @haoxiongliu on GitHub (Feb 5, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-R1/issues/282

Dear DeepSeek Team,
Hello! R1 is an execellent work and thank you for sharing the model checkpoints and techinical details in the report.

I have several questions here:

Could you share more information on the question set used in R1-Zero? We can only know that the training persists for 8000 steps from the report, while the group size and the total number of questions remain unclear. Also, It would be kind of you to share the criteria you select the problems.
Have you tried offline/online RFT counterparts of R1-Zero? I am curious with the performance gap between different training methods. Preliminaries results on small models are also appreciable.
In R1 training, how strong is the model after cold-start? Could you share some evaluation results?

I would greatly appreciate any insights you could share on these matters. Thank you again for your valuable contributions to the open-source community.

Originally created by @haoxiongliu on GitHub (Feb 5, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-R1/issues/282 Dear DeepSeek Team, Hello! R1 is an execellent work and thank you for sharing the model checkpoints and techinical details in the report. I have several questions here: 1. Could you share more information on the question set used in R1-Zero? We can only know that the training persists for 8000 steps from the report, while the group size and the total number of questions remain unclear. Also, It would be kind of you to share the criteria you select the problems. 2. Have you tried offline/online RFT counterparts of R1-Zero? I am curious with the performance gap between different training methods. Preliminaries results on small models are also appreciable. 3. In R1 training, how strong is the model after cold-start? Could you share some evaluation results? I would greatly appreciate any insights you could share on these matters. Thank you again for your valuable contributions to the open-source community.

bsakaguchi

2026-10-26 21:16:56 +00:00

closed this issue
added the
stale

closed-as-stale
labels

bsakaguchi commented

2026-10-26 21:16:58 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (Mar 30, 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 30, 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 21:16:58 +00:00

Author

Owner

@github-actions[bot] commented on GitHub (May 6, 2025):

false

@github-actions[bot] commented on GitHub (May 6, 2025): false

bsakaguchi referenced this issue

2026-10-26 21:26:37 +00:00

[PR #213] [CLOSED] docs: update README.md #773