SocialMind

Model track

Model submission page

Back to introduction

Model track

Submission details

Checking sign-in and team access... Go to team page
  • Open-weight model, no more than 20B parameters.
  • Revision defaults to master. Fixed tags and commit hashes are recommended; main and latest are rejected.
  • Use a time-limited read-only token; it is deleted after model caching.
  • Before submitting, confirm that the model starts successfully with Transformers 4.51.3 and vLLM 0.8.5.post1 using vllm serve. Models that fail to start cannot be evaluated.
  • Stage A and Stage B quotas are separate. Stage A allows up to 2 completed evaluations per team per Beijing calendar day; Stage B allows up to 3 completed evaluations per team for the entire stage. Queued or running evaluations reserve the corresponding slot, and failures release it. After leaving a team, participants cannot create or join another team for 24 hours.
  • Queue protection: each account may send up to 12 submission requests per 60 minutes. This does not consume the daily completed-evaluation quota.
Loading team... —
The evaluation worker loads the model from this ModelScope repository; the display name below is not used for loading.
? Go to https://modelscope.cn/my/access/token to get a token.
Token is deleted after model caching.
Revision defaults to master. Fixed tags and commit hashes are recommended; main and latest are rejected.
This name is shown in evaluation records and on the leaderboard only.
The worker verifies the actual parameter count after download. Models over 20B are stopped without consuming quota.