{
  "model_id": "Baragi-AI/Munche-v2-768",
  "base_model": "google/embeddinggemma-300m",
  "source_checkpoint": "outputs/style_embedding_v9_original768_proto4_from200/fold-0/checkpoint-600",
  "embedding_dimension": 768,
  "max_seq_length": 1024,
  "document_prompt": "title: none | text: ",
  "adapter_merged": false,
  "deployment_dtype": "bfloat16",
  "auxiliary_heads_included": false,
  "source_training_metadata": {
    "step": 600,
    "stage": "content_joint",
    "config": {
      "data": {
        "processed_root": "processed_data",
        "package_root": "processed_data/munche_style_datasets_v1",
        "registry_path": "artifacts/split_registry_selected_v1.jsonl",
        "stylometry_path": "artifacts/stylometry_windows_reliable24_v2.npz",
        "ai_stylometry_path": "artifacts/ai_stylometry_v1.npz",
        "semantic_cache_path": "artifacts/semantic_embeddings_v1.npz",
        "split_seed": 20260805,
        "train_ratio": 0.8,
        "validation_ratio": 0.1,
        "test_ratio": 0.1,
        "folds": 5,
        "fold": 0,
        "num_workers": 0
      },
      "model": {
        "base_model": "google/embeddinggemma-300m",
        "embedding_mode": "original_768",
        "document_prompt": "title: none | text: ",
        "style_dim": 768,
        "layer_mix_count": 4,
        "pooling_attention_dim": 256,
        "projection_dropout": 0.1,
        "lora_rank": 16,
        "lora_alpha": 32,
        "lora_dropout": 0.05,
        "lora_targets": [
          "q_proj",
          "v_proj",
          "o_proj"
        ],
        "publication_classes": 5,
        "publication_hidden_dim": 128,
        "stylometry_dim": 24,
        "stylometry_hidden_dim": 256,
        "dtype": "bfloat16",
        "gradient_checkpointing": false
      },
      "window": {
        "train_lengths": [
          512,
          768,
          1024
        ],
        "inference_length": 1024,
        "inference_stride": 512,
        "boundary_slack": 64,
        "windows_per_work_eval": 8
      },
      "loss": {
        "temperature": 0.07,
        "prototype_temperature": 0.07,
        "prototype_cluster_temperature": 0.1,
        "prototype_count": 3,
        "prototype_coverage": 0.15,
        "prototype_diversity": 0.1,
        "prototype_dominance": 0.0,
        "prototype_assignment_sharpness": 0.0,
        "prototype_usage": 0.0,
        "prototype_split_consistency": 0.0,
        "prototype_min_distance": 0.15,
        "prototype_min_works": 2.0,
        "prototype_min_work_mass": 0.75,
        "prototype_min_second_work_share": 0.2,
        "prototype_presence_temperature": 0.2,
        "prototype_max_work_dominance": 0.65,
        "prototype_min_usage": 0.15,
        "prototype_assignment_margin": 0.3,
        "ranking_margin": 0.15,
        "synthetic_margin": 0.08,
        "work": 0.8,
        "author": 0.5,
        "prototype": 0.6,
        "hard_negative": 0.25,
        "counterfactual": 0.75,
        "synthetic": 0.75,
        "decorrelation": 0.1,
        "stylometry": 0.5,
        "publication": 1.0,
        "publication_huber": 0.3,
        "publication_ranking": 0.3,
        "publication_cutpoint": 0.05,
        "publication_consistency": 0.0,
        "human_ai": 0.6
      },
      "train": {
        "output_dir": "outputs/style_embedding_v9_original768_proto4_from200/fold-0",
        "seed": 20260805,
        "warm_start_checkpoint": null,
        "resume_model_checkpoint": "outputs/style_embedding_v8_original768_r1/fold-0/checkpoint-200",
        "stages": [
          "warmup",
          "metric",
          "content_joint"
        ],
        "stage_steps": {
          "warmup": 0,
          "metric": 0,
          "content_joint": 800
        },
        "loss_ramp_steps": {
          "author": 0,
          "prototype": 0,
          "prototype_coverage": 0,
          "prototype_diversity": 0,
          "prototype_dominance": 0,
          "prototype_assignment_sharpness": 0,
          "prototype_usage": 0,
          "prototype_split_consistency": 0,
          "hard_negative": 100,
          "counterfactual": 100,
          "decorrelation": 100,
          "stylometry_ai": 150,
          "synthetic": 100,
          "human_ai": 150
        },
        "loss_fade_starts": {
          "work": 400,
          "author": 400,
          "prototype": 400,
          "stylometry": 100,
          "stylometry_ai": 300,
          "synthetic": 300,
          "decorrelation": 400
        },
        "loss_fade_steps": {
          "work": 400,
          "author": 400,
          "prototype": 400,
          "stylometry": 400,
          "stylometry_ai": 500,
          "synthetic": 500,
          "decorrelation": 100
        },
        "loss_fade_floors": {
          "work": 0.625,
          "author": 0.4,
          "prototype": 0.6666667,
          "stylometry": 0.1,
          "stylometry_ai": 0.1,
          "synthetic": 0.0,
          "decorrelation": 0.5
        },
        "human_metric_every": 2,
        "authors_per_batch": 8,
        "works_per_author": 3,
        "windows_per_work": 2,
        "prototype_batch_every": 10,
        "prototype_batch_offset": 5,
        "prototype_batch_authors": 4,
        "prototype_batch_works_per_author": 4,
        "prototype_batch_windows_per_work": 3,
        "prototype_batch_coverage_multiplier": 1.5,
        "extra_works_per_batch": 4,
        "auxiliary_batch_size": 8,
        "auxiliary_forward_batch_size": 4,
        "publication_every": 4,
        "publication_works_per_class": 2,
        "publication_interval_works": 4,
        "publication_windows_per_work": 4,
        "publication_temperature": 0.5,
        "publication_ranking_margin": 0.5,
        "publication_interval_weight": 0.25,
        "publication_max_interval_width": 3,
        "publication_encoder_gradient_scale": 0.25,
        "publication_head_only_updates": 0,
        "publication_warmup_ratio": 0.02,
        "human_ai_encoder_gradient_scale": 0.3,
        "decorrelation_guardrail_start": 400,
        "decorrelation_guardrail_target": 0.18,
        "gradient_accumulation": 1,
        "head_learning_rate": 8e-05,
        "publication_head_learning_rate": 0.0002,
        "human_ai_head_learning_rate": 0.0002,
        "lora_learning_rate": 2e-05,
        "weight_decay": 0.01,
        "warmup_ratio": 0.02,
        "max_grad_norm": 50.0,
        "log_every": 20,
        "save_every": 200,
        "semantic_hard_k": 20,
        "validation_full_every": 200,
        "validation_encode_batch_size": 64,
        "validation_prototype_support_works": 2,
        "validation_prototype_windows_per_work": 6,
        "validation_prototype_min_windows_per_work": 3,
        "validation_prototype_count": 3,
        "validation_human_ai_samples_per_source": 16,
        "wandb_project": "munche-style-embedding",
        "wandb_run_name": "style-v9-original768-proto4-from-v8-200-plus800",
        "wandb_mode": "online"
      }
    }
  }
}