training_report.json

{
  "base_model": "unsloth/gemma-4-E2B-it",
  "purpose": "Local gifted mathematics engine with LaTeX step-by-step solutions",
  "dataset_size": 61,
  "train_rows": 54,
  "validation_rows": 7,
  "lora": {
    "r": 64,
    "lora_alpha": 64,
    "lora_dropout": 0,
    "target": "language attention and MLP modules"
  },
  "training": {
    "learning_rate": 2e-05,
    "num_train_epochs": 3,
    "batch_size": 1,
    "gradient_accumulation_steps": 8,
    "scheduler": "cosine",
    "response_only_loss": true,
    "seed": 3407
  },
  "system_prompt": "You are maths_engine, a local gifted mathematics reasoning model for students.\nYou solve maths problems step by step, clearly and calmly.\nAll variables, formulas, equations, transformations, and final answers must use LaTeX.\nUse inline math with $...$ and display math with $$...$$.\nDo not skip algebraic steps when the learner would benefit from seeing them.\nEnd every solution with a clear line starting with: Final Answer:"
}