training_results_grpo_3b_v3.json

{
  "env_url": "https://shasidharyadavr-netweaver-sre.hf.space",
  "model_name": "Qwen/Qwen2.5-3B-Instruct",
  "training_method": "GRPO v3: 4-bit + LoRA r=64 (q_proj+k_proj+v_proj+o_proj+gate_proj+up_proj+down_proj), LR=5e-05, temp_train=temp_eval=0.8",
  "max_train_steps": 40,
  "num_generations": 4,
  "learning_rate": 5e-05,
  "lora_rank": 64,
  "lora_target_modules": [
    "q_proj",
    "k_proj",
    "v_proj",
    "o_proj",
    "gate_proj",
    "up_proj",
    "down_proj"
  ],
  "training_temperature": 0.8,
  "eval_temperature": 0.8,
  "trainable_params": 119734272,
  "total_params": 1818406912,
  "training_rewards": [
    0.999,
    0.88,
    0.88,
    0.88,
    0.999,
    0.86,
    0.5,
    0.86,
    0.4,
    0.667,
    0.667,
    0.667,
    0.3,
    0.86,
    0.3,
    0.86,
    0.84,
    0.84,
    0.84,
    0.84,
    0.667,
    0.667,
    0.467,
    0.467,
    0.84,
    0.999,
    0.84,
    0.84,
    0.5,
    0.5,
    0.3,
    0.5,
    0.3,
    0.3,
    0.86,
    0.86,
    0.84,
    0.5,
    0.84,
    0.84
  ],
  "before_rewards": [
    0.999,
    0.733,
    0.9,
    0.9,
    0.667,
    0.9,
    0.999,
    0.999,
    0.5,
    0.667
  ],
  "after_rewards": [
    0.4,
    0.3,
    0.3,
    0.3,
    0.3,
    0.3,
    0.3,
    0.3,
    0.4,
    0.3
  ],
  "difficulty_breakdown": {
    "before": {
      "easy": 0.9,
      "medium": 0.87425,
      "hard": 0.7734
    },
    "after": {
      "easy": 0.3,
      "medium": 0.3,
      "hard": 0.3666666666666667
    }
  },
  "timestamp": "2026-04-26 02:23:05",
  "source": "grpo_3b_real_run_v3",
  "job_flavor": "a10g-small",
  "notes": "GRPO v3 on Qwen/Qwen2.5-3B-Instruct (4-bit + LoRA r=64, targets=['q_proj', 'k_proj', 'v_proj', 'o_proj', 'gate_proj', 'up_proj', 'down_proj']) vs the live NetWeaver SRE env. Train temp = eval temp = 0.8, LR = 5e-05. Heuristic baseline preserved (separate filenames)."
}