a paper I discussed above on membership inference attacks.
The New Cosmopolitanism
，更多细节参见新收录的资料
CodeforcesThe coding capabilities of Sarvam 30B and Sarvam 105B were evaluated using real-world competitive programming problems from Codeforces (Div3, link). The evaluation involved generating Python solutions and manually submitting them to the Codeforces platform to verify correctness. Correctness is measured at pass@1 and pass@4 as shown in the table below.
2026-03-09 22:00:00