metadata.json

{
  "model_name": "tat-llm-final-e4",
  "base_model": "NousResearch/Nous-Hermes-2-Mistral-7B-DPO",
  "tokenizer": "NousResearch/Nous-Hermes-2-Mistral-7B-DPO",
  "adapter_type": "LoRA",
  "adapter_config": {
    "r": 64,
    "alpha": 16,
    "dropout": 0.1,
    "bias": "none"
  },
  "training": {
    "dataset": "TAT-QA (train.json)",
    "num_examples": 9000,
    "num_epochs": 4,
    "max_seq_length": 1024,
    "batch_size_per_device": 2,
    "learning_rate": 0.0002,
    "lr_scheduler": "constant",
    "fp16": false,
    "bf16": true,
    "optimizer": "AdamW (via Trainer)"
  },
  "notes": "Instruction-tuned with simplified prompt format. No evaluation run due to memory constraints. Use .generate() for inference.",
  "created_by": "Your Name or Team",
  "date": "2025-07-08"
}