{
  "release_name": "Qwen3.6-27B-Thinking-SecOPD",
  "base_model": "Qwen/Qwen3.6-27B",
  "paper_model": "SecOPD",
  "checkpoint_step": 150,
  "weights": {
    "format": "safetensors",
    "merged_lora": true,
    "shards": 15,
    "tensor_entries": 1199,
    "indexed_total_size_bytes": 55562855904
  },
  "training": {
    "recipe": "clean-context on-policy distillation",
    "lora_rank": 128,
    "learning_rate": 0.0001,
    "sampling_temperature": 1.0,
    "maximum_generation_tokens": 16384
  },
  "tokenizer": {
    "vocabulary_size": 248077,
    "input_role_template": true,
    "note": "The vocabulary and token IDs match the merged checkpoint tokenizer exactly. The chat template is the input-role template used by the paper evaluation."
  },
  "code": "https://github.com/pppyb/SecOPD"
}