{
 "base_model": "Qwen/Qwen3.5-2B",
 "lora_rank": 32,
 "renderer": "qwen3_5_disable_thinking",
 "arms": {
  "grpo_gs8_seed0": {
   "group_size": 8,
   "groups_per_batch": 8
  },
  "gs1_seed0": {
   "group_size": 1,
   "groups_per_batch": 64,
   "kernel": "word TF-IDF cosine (sklearn defaults, IDF fit on 7473 train questions), w=exp(cos/0.1), self excluded, normalized"
  }
 },
 "steps": 40,
 "pilot_steps": 3,
 "batch_size": 64,
 "max_tokens": 512,
 "max_seq_len": 1024,
 "temperature": 1.0,
 "loss_fn": "importance_sampling",
 "optimizer": {
  "lr": 4e-05,
  "beta1": 0.9,
  "beta2": 0.95,
  "eps": 1e-08
 },
 "dataset": "openai/gsm8k main train (shuffle seed 0)",
 "n_eval": 1319,
 "eval_decoding": {
  "temperature": 0,
  "max_tokens": 512
 },
 "tinker": "0.37.0",
 "tinker_cookbook": "6daabe0838736557f0e802020b710073eb340287",
 "init": "create_lora_training_client(rank=32) [seed rejected by server] -> save_state -> create_training_client_from_state per arm"
}