Commit graph

12 commits

Author SHA1 Message Date
balyan.sid@gmail.com
5a20abdce7 switch eval to use managed server adapter impl. moved managed server
adapter
2026-01-23 23:26:29 +05:30
balyan.sid@gmail.com
6a27e88023 use managed server 2026-01-14 17:09:01 +05:30
balyan.sid@gmail.com
32320512e8 update verifiers_server to use tokenizer_for_trainer 2026-01-13 15:00:54 +05:30
balyan.sid@gmail.com
a1d1e7d7fe fix env_args, dataset/prompt loading 2026-01-12 10:39:43 +05:30
balyan.sid@gmail.com
9db6c0d1ed added better wandb logging 2026-01-12 10:34:05 +05:30
balyan.sid@gmail.com
dceb1d8fd8 parallelize verifiers_server: use generate() for SFT, parallel
ManagedServer contexts for RL
2026-01-12 10:34:05 +05:30
balyan.sid@gmail.com
24b4488c60 clean up eval, pin verifiers version 2026-01-12 10:34:05 +05:30
balyan.sid@gmail.com
cf636595d2 rework server and eval for rl rollout. add in asyncmanagedserver for
verifiers
2026-01-12 10:34:05 +05:30
balyan.sid@gmail.com
5b09ad86f4 update readme, add sft-datagen to verifiers_server 2026-01-09 19:20:41 +05:30
balyan.sid@gmail.com
dda85430da fix docstrings 2026-01-09 16:25:44 +05:30
balyan.sid@gmail.com
9d5cd2b593 fix: improve verifiers environments consistency and correctness
- verifiers_server.py: consistent dataset column selection for train/test,
  remove redundant comments, preserve float precision for scores
- verifiers_eval.py: add env_config_cls, fix constructor signature to match
  BaseEnv (slurm bool), make stub methods raise NotImplementedError
2026-01-09 16:21:12 +05:30
balyan.sid@gmail.com
ed826de724 wip: verifiers integration 2026-01-09 14:21:03 +05:30