Upload completed TypePredictor run 2026-07-12T18:56:38+00:00

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+predictions/test_predictions.jsonl filter=lfs diff=lfs merge=lfs -text
+predictions/validation_predictions.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,136 @@
+---
+language:
+- ar
+library_name: pytorch
+pipeline_tag: text-classification
+tags:
+- arabic
+- named-entity-recognition
+- entity-typing
+- wojood
+- neoarabert
+datasets:
+- U4RASD/TypePrediction
+metrics:
+- accuracy
+- f1
+---
+
+# TypePredictor
+
+`TypePredictor` is a mention-level Arabic entity-type classifier trained on
+`U4RASD/TypePrediction`. Given an Arabic sentence and a known character span, the
+span is wrapped with generic `[ENT]` and `[/ENT]` markers and classified into one
+of 21 coarse Wojood entity types.
+
+> This README is the automatically generated preliminary model card. The RunPod
+> package instructs Codex to wait for Ahmad's explicit confirmation that the full
+> training, validation, and test workflow is complete before writing and pushing
+> the final polished model card from all run artifacts.
+
+## Architecture
+
+```text
+sentence + known span
+-> insert [ENT] ... [/ENT]
+-> U4RASD/NeoAraBERT
+-> CLS vector [768]
+-> Dropout(0.1)
+-> Linear(768, 21)
+-> argmax entity type
+```
+
+The baseline uses ordinary unweighted multiclass cross-entropy. There is no
+class weighting, weighted sampler, extra MLP, span pooling, threshold, or
+subject/object-specific model.
+
+## Labels
+
+GPE, ORG, DATE, PERS, NORP, ORDINAL, OCC, EVENT, CARDINAL, LOC, WEBSITE, FAC, LAW, TIME, MONEY, CURR, LANGUAGE, PERCENT, PRODUCT, QUANTITY, UNIT
+
+## Dataset
+
+- Train: 100,796
+- Validation: 12,600
+- Test: 12,600
+- Validation seen-sentence/new-entity: 11,661
+- Validation unseen-sentence: 939
+- Test seen-sentence/new-entity: 11,606
+- Test unseen-sentence: 994
+
+The split is mention-level and intentionally allows the same sentence to appear
+across splits with different target mentions. Category-specific results should
+therefore be interpreted separately.
+
+## Validation results
+
+| Subset | Rows | Accuracy | Micro F1 | Macro F1 (present types) | Weighted F1 |
+|---|---:|---:|---:|---:|---:|
+| overall | 12600 | 0.977857 | 0.977857 | 0.958728 | 0.977791 |
+| unseen_sentence | 939 | 0.945687 | 0.945687 | 0.908841 | 0.945011 |
+| seen_sentence_new_entity | 11661 | 0.980448 | 0.980448 | 0.963262 | 0.980403 |
+
+## Test results
+
+| Subset | Rows | Accuracy | Micro F1 | Macro F1 (present types) | Weighted F1 |
+|---|---:|---:|---:|---:|---:|
+| overall | 12600 | 0.979365 | 0.979365 | 0.959920 | 0.979297 |
+| unseen_sentence | 994 | 0.962777 | 0.962777 | 0.893174 | 0.961780 |
+| seen_sentence_new_entity | 11606 | 0.980786 | 0.980786 | 0.964113 | 0.980712 |
+
+Detailed per-class reports, confusion matrices, predictions, configuration, and
+checkpoint metadata are included in the repository artifacts.
+
+## Checkpoints
+
+- Best checkpoint: epoch `3.9682923925552602`, step `25000`
+- Best validation macro F1: `0.958728`
+- Latest checkpoint: epoch `3.9994047382832654`, step `25196`
+- Repository root: best model
+- `checkpoints/latest/`: latest completed checkpoint snapshot
+
+Both the best and latest full checkpoints are also preserved in the local RunPod
+output directory.
+
+## Training configuration
+
+- Encoder learning rate: `1e-05`
+- Classifier learning rate: `5e-05`
+- Epochs: `4.0`
+- Train batch size: `4`
+- Gradient accumulation: `4`
+- Effective batch size: `16`
+- Maximum length: `512`
+- FP16: `True`
+- Best-checkpoint criterion: validation macro F1
+
+## Loading
+
+The repository contains `modeling_type_predictor.py` because this classifier is
+a small custom PyTorch wrapper around NeoAraBERT.
+
+```python
+from huggingface_hub import hf_hub_download
+import importlib.util
+
+source = hf_hub_download("U4RASD/TypePredictor", "modeling_type_predictor.py")
+spec = importlib.util.spec_from_file_location("modeling_type_predictor", source)
+module = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(module)
+model, tokenizer, config = module.NeoAraBERTTypePredictor.from_pretrained(
+    "U4RASD/TypePredictor"
+)
+```
+
+## Intended use
+
+Use the model when the target entity mention and its character span are already
+known. The same model can classify a relation subject or object mention.
+
+## Limitations
+
+- The model does not detect spans; it classifies a supplied span.
+- The split contains sentence overlap by design, so overall metrics are not a
+  pure unseen-sentence estimate.
+- Rare classes such as `UNIT` and `QUANTITY` have limited support.
+- This first experiment intentionally uses no imbalance correction.

checkpoint_summary.json

@@ -0,0 +1,13 @@
+{
+  "best_checkpoint_source": "outputs/TypePredictor/checkpoints/checkpoint-25000",
+  "best_checkpoint_preserved": "outputs/TypePredictor/best_checkpoint",
+  "best_step": 25000,
+  "best_epoch": 3.9682923925552602,
+  "best_validation_macro_f1": 0.9587281193044483,
+  "latest_checkpoint_source": "outputs/TypePredictor/checkpoints/checkpoint-25196",
+  "latest_checkpoint_preserved": "outputs/TypePredictor/latest_checkpoint",
+  "latest_step": 25196,
+  "latest_epoch": 3.9994047382832654,
+  "best_and_latest_same_source": false,
+  "root_released_model": "best checkpoint"
+}

checkpoints/latest/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f50dc9dafa68f1d1e36c152f307cd243c56b3f4a76f1e8a88bbaf64fe8becbf6
+size 1985443322

checkpoints/latest/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3ef5bb9771e8705208a0f3e97788305b09b19d0be51fcf939f736fc5c13fb72
+size 992721978

checkpoints/latest/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:923e5b6a51442e1617111f28197dff670cd469df9e1de7d6216bb14030444691
+size 14244

checkpoints/latest/scaler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cfb01e00465a5891d4570d1c7db65c31941253c42933e7c82e41ca335ed13f5
+size 988

checkpoints/latest/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caa8aa904b6ba57c95fdeb6646034707f9aa8f50f77bbe64ebde1dfff45a23e3
+size 1064

checkpoints/latest/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5b6cfce7351bff5319a51e160874bfbfcfd153324040fb99c38f5fff0732e17
+size 5368

config.json

@@ -0,0 +1,102 @@
+{
+  "architecture": "NeoAraBERTTypePredictor",
+  "task": "mention-level 21-class Arabic entity typing",
+  "base_model": "U4RASD/NeoAraBERT",
+  "input_markers": [
+    "[ENT]",
+    "[/ENT]"
+  ],
+  "representation": "last hidden state at token position 0 (CLS)",
+  "dropout": 0.1,
+  "classifier": [
+    768,
+    21
+  ],
+  "loss": "ordinary unweighted multiclass cross-entropy",
+  "num_labels": 21,
+  "labels": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "tokenizer_size": 65002,
+  "hidden_size": 768,
+  "max_length": 512,
+  "context_candidates": [
+    null,
+    500,
+    300,
+    150,
+    80,
+    30,
+    0
+  ],
+  "architectures": [
+    "NeoAraBERTTypePredictor"
+  ],
+  "model_type": "type_predictor",
+  "id2label": {
+    "0": "GPE",
+    "1": "ORG",
+    "2": "DATE",
+    "3": "PERS",
+    "4": "NORP",
+    "5": "ORDINAL",
+    "6": "OCC",
+    "7": "EVENT",
+    "8": "CARDINAL",
+    "9": "LOC",
+    "10": "WEBSITE",
+    "11": "FAC",
+    "12": "LAW",
+    "13": "TIME",
+    "14": "MONEY",
+    "15": "CURR",
+    "16": "LANGUAGE",
+    "17": "PERCENT",
+    "18": "PRODUCT",
+    "19": "QUANTITY",
+    "20": "UNIT"
+  },
+  "label2id": {
+    "GPE": 0,
+    "ORG": 1,
+    "DATE": 2,
+    "PERS": 3,
+    "NORP": 4,
+    "ORDINAL": 5,
+    "OCC": 6,
+    "EVENT": 7,
+    "CARDINAL": 8,
+    "LOC": 9,
+    "WEBSITE": 10,
+    "FAC": 11,
+    "LAW": 12,
+    "TIME": 13,
+    "MONEY": 14,
+    "CURR": 15,
+    "LANGUAGE": 16,
+    "PERCENT": 17,
+    "PRODUCT": 18,
+    "QUANTITY": 19,
+    "UNIT": 20
+  }
+}

configs/architecture_config.json

@@ -0,0 +1,52 @@
+{
+  "architecture": "NeoAraBERTTypePredictor",
+  "task": "mention-level 21-class Arabic entity typing",
+  "base_model": "U4RASD/NeoAraBERT",
+  "input_markers": [
+    "[ENT]",
+    "[/ENT]"
+  ],
+  "representation": "last hidden state at token position 0 (CLS)",
+  "dropout": 0.1,
+  "classifier": [
+    768,
+    21
+  ],
+  "loss": "ordinary unweighted multiclass cross-entropy",
+  "num_labels": 21,
+  "labels": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "tokenizer_size": 65002,
+  "hidden_size": 768,
+  "max_length": 512,
+  "context_candidates": [
+    null,
+    500,
+    300,
+    150,
+    80,
+    30,
+    0
+  ]
+}

configs/id2label.json

@@ -0,0 +1,23 @@
+{
+  "0": "GPE",
+  "1": "ORG",
+  "2": "DATE",
+  "3": "PERS",
+  "4": "NORP",
+  "5": "ORDINAL",
+  "6": "OCC",
+  "7": "EVENT",
+  "8": "CARDINAL",
+  "9": "LOC",
+  "10": "WEBSITE",
+  "11": "FAC",
+  "12": "LAW",
+  "13": "TIME",
+  "14": "MONEY",
+  "15": "CURR",
+  "16": "LANGUAGE",
+  "17": "PERCENT",
+  "18": "PRODUCT",
+  "19": "QUANTITY",
+  "20": "UNIT"
+}

configs/label2id.json

@@ -0,0 +1,23 @@
+{
+  "GPE": 0,
+  "ORG": 1,
+  "DATE": 2,
+  "PERS": 3,
+  "NORP": 4,
+  "ORDINAL": 5,
+  "OCC": 6,
+  "EVENT": 7,
+  "CARDINAL": 8,
+  "LOC": 9,
+  "WEBSITE": 10,
+  "FAC": 11,
+  "LAW": 12,
+  "TIME": 13,
+  "MONEY": 14,
+  "CURR": 15,
+  "LANGUAGE": 16,
+  "PERCENT": 17,
+  "PRODUCT": 18,
+  "QUANTITY": 19,
+  "UNIT": 20
+}

configs/run_config.json

@@ -0,0 +1,76 @@
+{
+  "experiment_name": "TypePredictor",
+  "dataset_repo_id": "U4RASD/TypePrediction",
+  "dataset_files": {
+    "train": "type_predictor_train.jsonl",
+    "validation": "type_predictor_val.jsonl",
+    "test": "type_predictor_test.jsonl"
+  },
+  "model_name": "U4RASD/NeoAraBERT",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "seed": 42,
+  "max_length": 512,
+  "context_chars": 300,
+  "context_candidates": [
+    null,
+    500,
+    300,
+    150,
+    80,
+    30,
+    0
+  ],
+  "num_epochs": 4.0,
+  "train_batch_size": 4,
+  "eval_batch_size": 8,
+  "gradient_accumulation_steps": 4,
+  "effective_batch_size": 16,
+  "encoder_lr": 1e-05,
+  "head_lr": 5e-05,
+  "dropout": 0.1,
+  "weight_decay": 0.01,
+  "warmup_ratio": 0.1,
+  "max_grad_norm": 1.0,
+  "logging_steps": 50,
+  "eval_steps": 500,
+  "save_steps": 500,
+  "fp16": true,
+  "best_metric_name": "validation macro F1",
+  "parameter_counts": {
+    "total": 248162325,
+    "trainable": 248162325,
+    "encoder": 248146176,
+    "classifier": 16149
+  },
+  "checkpoint_summary": {
+    "best_checkpoint_source": "outputs/TypePredictor/checkpoints/checkpoint-25000",
+    "best_checkpoint_preserved": "outputs/TypePredictor/best_checkpoint",
+    "best_step": 25000,
+    "best_epoch": 3.9682923925552602,
+    "best_validation_macro_f1": 0.9587281193044483,
+    "latest_checkpoint_source": "outputs/TypePredictor/checkpoints/checkpoint-25196",
+    "latest_checkpoint_preserved": "outputs/TypePredictor/latest_checkpoint",
+    "latest_step": 25196,
+    "latest_epoch": 3.9994047382832654,
+    "best_and_latest_same_source": false,
+    "root_released_model": "best checkpoint"
+  },
+  "training_started_at_utc": "2026-07-12T15:30:38+00:00",
+  "training_finished_at_utc": "2026-07-12T18:55:01+00:00",
+  "training_wall_seconds": 12262.467122793198,
+  "versions": {
+    "python": "3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0]",
+    "torch": "2.5.1+cu124",
+    "transformers": "4.49.0",
+    "huggingface_hub": "0.29.1",
+    "scikit_learn": "1.5.2",
+    "numpy": "2.0.2",
+    "pandas": "2.2.3",
+    "platform": "Linux-6.8.0-48-generic-x86_64-with-glibc2.39"
+  },
+  "hardware": {
+    "cuda_available": true,
+    "gpu_name": "NVIDIA A40",
+    "gpu_vram_gb": 44.431640625
+  }
+}

configs/training_arguments.json

@@ -0,0 +1,145 @@
+{
+  "output_dir": "outputs/TypePredictor/checkpoints",
+  "overwrite_output_dir": false,
+  "do_train": false,
+  "do_eval": true,
+  "do_predict": false,
+  "eval_strategy": "steps",
+  "prediction_loss_only": false,
+  "per_device_train_batch_size": 4,
+  "per_device_eval_batch_size": 8,
+  "per_gpu_train_batch_size": null,
+  "per_gpu_eval_batch_size": null,
+  "gradient_accumulation_steps": 4,
+  "eval_accumulation_steps": null,
+  "eval_delay": 0,
+  "torch_empty_cache_steps": null,
+  "learning_rate": 5e-05,
+  "weight_decay": 0.01,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "adam_epsilon": 1e-08,
+  "max_grad_norm": 1.0,
+  "num_train_epochs": 4.0,
+  "max_steps": -1,
+  "lr_scheduler_type": "linear",
+  "lr_scheduler_kwargs": {},
+  "warmup_ratio": 0.1,
+  "warmup_steps": 0,
+  "log_level": "passive",
+  "log_level_replica": "warning",
+  "log_on_each_node": true,
+  "logging_dir": "outputs/TypePredictor/checkpoints/runs/Jul12_15-30-37_547e2be95e65",
+  "logging_strategy": "steps",
+  "logging_first_step": false,
+  "logging_steps": 50,
+  "logging_nan_inf_filter": true,
+  "save_strategy": "steps",
+  "save_steps": 500,
+  "save_total_limit": 2,
+  "save_safetensors": false,
+  "save_on_each_node": false,
+  "save_only_model": false,
+  "restore_callback_states_from_checkpoint": false,
+  "no_cuda": false,
+  "use_cpu": false,
+  "use_mps_device": false,
+  "seed": 42,
+  "data_seed": 42,
+  "jit_mode_eval": false,
+  "use_ipex": false,
+  "bf16": false,
+  "fp16": true,
+  "fp16_opt_level": "O1",
+  "half_precision_backend": "auto",
+  "bf16_full_eval": false,
+  "fp16_full_eval": false,
+  "tf32": null,
+  "local_rank": 0,
+  "ddp_backend": null,
+  "tpu_num_cores": null,
+  "tpu_metrics_debug": false,
+  "debug": [],
+  "dataloader_drop_last": false,
+  "eval_steps": 500,
+  "dataloader_num_workers": 2,
+  "dataloader_prefetch_factor": null,
+  "past_index": -1,
+  "run_name": "outputs/TypePredictor/checkpoints",
+  "disable_tqdm": false,
+  "remove_unused_columns": false,
+  "label_names": null,
+  "load_best_model_at_end": true,
+  "metric_for_best_model": "macro_f1",
+  "greater_is_better": true,
+  "ignore_data_skip": false,
+  "fsdp": [],
+  "fsdp_min_num_params": 0,
+  "fsdp_config": {
+    "min_num_params": 0,
+    "xla": false,
+    "xla_fsdp_v2": false,
+    "xla_fsdp_grad_ckpt": false
+  },
+  "fsdp_transformer_layer_cls_to_wrap": null,
+  "accelerator_config": {
+    "split_batches": false,
+    "dispatch_batches": null,
+    "even_batches": true,
+    "use_seedable_sampler": true,
+    "non_blocking": false,
+    "gradient_accumulation_kwargs": null
+  },
+  "deepspeed": null,
+  "label_smoothing_factor": 0.0,
+  "optim": "adamw_torch",
+  "optim_args": null,
+  "adafactor": false,
+  "group_by_length": false,
+  "length_column_name": "length",
+  "report_to": [],
+  "ddp_find_unused_parameters": null,
+  "ddp_bucket_cap_mb": null,
+  "ddp_broadcast_buffers": null,
+  "dataloader_pin_memory": true,
+  "dataloader_persistent_workers": false,
+  "skip_memory_metrics": true,
+  "use_legacy_prediction_loop": false,
+  "push_to_hub": false,
+  "resume_from_checkpoint": null,
+  "hub_model_id": null,
+  "hub_strategy": "every_save",
+  "hub_token": "<HUB_TOKEN>",
+  "hub_private_repo": null,
+  "hub_always_push": false,
+  "gradient_checkpointing": false,
+  "gradient_checkpointing_kwargs": null,
+  "include_inputs_for_metrics": false,
+  "include_for_metrics": [],
+  "eval_do_concat_batches": true,
+  "fp16_backend": "auto",
+  "evaluation_strategy": null,
+  "push_to_hub_model_id": null,
+  "push_to_hub_organization": null,
+  "push_to_hub_token": "<PUSH_TO_HUB_TOKEN>",
+  "mp_parameters": "",
+  "auto_find_batch_size": false,
+  "full_determinism": false,
+  "torchdynamo": null,
+  "ray_scope": "last",
+  "ddp_timeout": 1800,
+  "torch_compile": false,
+  "torch_compile_backend": null,
+  "torch_compile_mode": null,
+  "dispatch_batches": null,
+  "split_batches": null,
+  "include_tokens_per_second": false,
+  "include_num_input_tokens_seen": false,
+  "neftune_noise_alpha": null,
+  "optim_target_modules": null,
+  "batch_eval_metrics": false,
+  "eval_on_start": false,
+  "use_liger_kernel": false,
+  "eval_use_gather_object": false,
+  "average_tokens_across_devices": false
+}

confusion_matrices/test_overall_confusion_matrix.csv

@@ -0,0 +1,22 @@
+,GPE,ORG,DATE,PERS,NORP,ORDINAL,OCC,EVENT,CARDINAL,LOC,WEBSITE,FAC,LAW,TIME,MONEY,CURR,LANGUAGE,PERCENT,PRODUCT,QUANTITY,UNIT
+GPE,2649,21,4,2,6,0,1,1,1,3,2,1,0,0,0,0,0,0,0,0,0
+ORG,36,2601,1,5,6,0,1,3,0,2,1,3,0,0,0,0,0,0,0,0,0
+DATE,2,0,1968,2,1,0,0,0,0,1,0,0,0,2,0,0,1,0,0,0,0
+PERS,4,4,2,1053,2,1,4,0,0,0,1,0,0,0,0,0,0,0,0,0,0
+NORP,3,7,1,5,999,0,1,1,1,0,0,1,0,1,0,0,1,0,0,0,0
+ORDINAL,0,5,3,1,1,768,0,0,2,0,0,0,0,1,0,0,0,0,0,0,0
+OCC,1,3,0,1,1,0,762,0,0,0,1,0,0,1,0,0,0,0,0,0,0
+EVENT,1,8,0,1,2,0,0,363,0,0,0,2,0,0,0,0,0,0,0,0,0
+CARDINAL,1,0,4,1,1,8,3,0,354,0,0,0,0,0,0,0,0,0,0,0,0
+LOC,11,2,0,1,2,0,0,0,0,223,0,0,0,0,0,0,0,0,0,0,0
+WEBSITE,0,1,0,0,0,0,0,0,0,0,146,0,0,1,0,0,0,0,0,0,0
+FAC,9,0,1,2,0,0,0,0,0,1,0,116,0,0,0,0,0,0,0,0,0
+LAW,0,0,0,0,0,0,0,0,0,0,0,0,91,0,0,0,0,0,0,0,0
+TIME,2,0,1,0,0,4,0,0,0,0,0,2,0,76,0,0,2,0,0,0,0
+MONEY,0,0,1,0,0,0,0,0,1,0,0,0,0,0,37,1,0,1,0,1,0
+CURR,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,40,0,0,0,0,0
+LANGUAGE,1,0,0,1,2,0,0,0,0,0,0,0,0,1,0,0,28,0,0,0,0
+PERCENT,0,0,0,0,0,0,0,0,3,0,0,0,0,0,1,0,0,27,0,0,0
+PRODUCT,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,18,0,0
+QUANTITY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,10,0
+UNIT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,11

confusion_matrices/test_seen_sentence_new_entity_confusion_matrix.csv

@@ -0,0 +1,22 @@
+,GPE,ORG,DATE,PERS,NORP,ORDINAL,OCC,EVENT,CARDINAL,LOC,WEBSITE,FAC,LAW,TIME,MONEY,CURR,LANGUAGE,PERCENT,PRODUCT,QUANTITY,UNIT
+GPE,2506,18,3,1,5,0,1,1,0,2,0,1,0,0,0,0,0,0,0,0,0
+ORG,33,2467,1,5,6,0,1,3,0,2,0,2,0,0,0,0,0,0,0,0,0
+DATE,2,0,1849,2,0,0,0,0,0,1,0,0,0,2,0,0,1,0,0,0,0
+PERS,2,3,1,914,2,1,4,0,0,0,1,0,0,0,0,0,0,0,0,0,0
+NORP,2,6,0,5,878,0,1,1,1,0,0,1,0,1,0,0,0,0,0,0,0
+ORDINAL,0,5,3,1,1,686,0,0,2,0,0,0,0,1,0,0,0,0,0,0,0
+OCC,1,3,0,1,0,0,714,0,0,0,1,0,0,0,0,0,0,0,0,0,0
+EVENT,1,8,0,0,2,0,0,350,0,0,0,2,0,0,0,0,0,0,0,0,0
+CARDINAL,1,0,4,1,1,6,3,0,307,0,0,0,0,0,0,0,0,0,0,0,0
+LOC,10,2,0,1,2,0,0,0,0,209,0,0,0,0,0,0,0,0,0,0,0
+WEBSITE,0,1,0,0,0,0,0,0,0,0,94,0,0,1,0,0,0,0,0,0,0
+FAC,9,0,1,2,0,0,0,0,0,1,0,108,0,0,0,0,0,0,0,0,0
+LAW,0,0,0,0,0,0,0,0,0,0,0,0,83,0,0,0,0,0,0,0,0
+TIME,2,0,1,0,0,4,0,0,0,0,0,2,0,58,0,0,1,0,0,0,0
+MONEY,0,0,0,0,0,0,0,0,1,0,0,0,0,0,37,1,0,1,0,1,0
+CURR,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,38,0,0,0,0,0
+LANGUAGE,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,24,0,0,0,0
+PERCENT,0,0,0,0,0,0,0,0,3,0,0,0,0,0,1,0,0,25,0,0,0
+PRODUCT,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,16,0,0
+QUANTITY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,0
+UNIT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,11

confusion_matrices/test_unseen_sentence_confusion_matrix.csv

@@ -0,0 +1,22 @@
+,GPE,ORG,DATE,PERS,NORP,ORDINAL,OCC,EVENT,CARDINAL,LOC,WEBSITE,FAC,LAW,TIME,MONEY,CURR,LANGUAGE,PERCENT,PRODUCT,QUANTITY,UNIT
+GPE,143,3,1,1,1,0,0,0,1,1,2,0,0,0,0,0,0,0,0,0,0
+ORG,3,134,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0
+DATE,0,0,119,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+PERS,2,1,1,139,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+NORP,1,1,1,0,121,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0
+ORDINAL,0,0,0,0,0,82,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+OCC,0,0,0,0,1,0,48,0,0,0,0,0,0,1,0,0,0,0,0,0,0
+EVENT,0,0,0,1,0,0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0
+CARDINAL,0,0,0,0,0,2,0,0,47,0,0,0,0,0,0,0,0,0,0,0,0
+LOC,1,0,0,0,0,0,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0
+WEBSITE,0,0,0,0,0,0,0,0,0,0,52,0,0,0,0,0,0,0,0,0,0
+FAC,0,0,0,0,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,0
+LAW,0,0,0,0,0,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0
+TIME,0,0,0,0,0,0,0,0,0,0,0,0,0,18,0,0,1,0,0,0,0
+MONEY,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+CURR,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0
+LANGUAGE,0,0,0,1,2,0,0,0,0,0,0,0,0,1,0,0,4,0,0,0,0
+PERCENT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0
+PRODUCT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0
+QUANTITY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0
+UNIT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0

confusion_matrices/validation_overall_confusion_matrix.csv

@@ -0,0 +1,22 @@
+,GPE,ORG,DATE,PERS,NORP,ORDINAL,OCC,EVENT,CARDINAL,LOC,WEBSITE,FAC,LAW,TIME,MONEY,CURR,LANGUAGE,PERCENT,PRODUCT,QUANTITY,UNIT
+GPE,2642,27,3,4,3,1,1,3,2,3,0,0,0,0,0,0,1,0,0,0,0
+ORG,21,2614,1,4,5,2,2,4,2,2,0,3,0,0,0,0,0,0,0,0,0
+DATE,3,0,1966,1,1,3,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0
+PERS,7,5,1,1042,7,1,5,0,0,0,1,3,0,0,0,0,0,0,0,0,0
+NORP,7,19,0,11,977,0,3,3,1,0,0,1,0,0,0,0,0,0,0,0,0
+ORDINAL,3,2,1,2,4,765,0,0,2,0,0,1,0,0,0,0,0,0,0,0,0
+OCC,2,1,0,5,1,0,759,0,0,0,1,0,0,0,0,0,0,0,0,0,0
+EVENT,4,10,2,0,2,0,1,355,1,0,0,2,0,0,0,0,0,0,0,0,0
+CARDINAL,0,0,1,0,0,6,0,0,365,0,0,0,0,1,0,0,0,0,0,0,0
+LOC,5,2,0,0,1,0,0,0,0,230,0,1,0,0,0,0,0,0,0,0,0
+WEBSITE,0,1,0,0,0,0,0,0,0,0,147,0,0,0,0,0,0,0,0,0,0
+FAC,6,2,0,0,0,0,0,1,0,0,0,121,0,0,0,0,0,0,0,0,0
+LAW,0,0,0,0,0,0,0,0,0,0,0,0,90,0,0,0,0,0,0,0,0
+TIME,2,1,4,0,0,2,0,0,0,1,0,0,0,76,0,0,1,0,0,0,0
+MONEY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,42,0,0,0,0,0,0
+CURR,0,0,0,0,0,0,1,0,0,0,0,0,0,0,1,39,0,0,0,0,0
+LANGUAGE,1,2,0,0,1,0,0,0,0,0,0,0,0,1,0,0,28,0,0,0,0
+PERCENT,0,0,0,0,0,0,0,0,2,0,0,0,0,0,1,0,0,28,0,0,0
+PRODUCT,0,1,0,1,0,1,0,0,0,0,0,0,0,0,0,0,0,0,16,0,0
+QUANTITY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,9,1
+UNIT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,10

confusion_matrices/validation_seen_sentence_new_entity_confusion_matrix.csv

@@ -0,0 +1,22 @@
+,GPE,ORG,DATE,PERS,NORP,ORDINAL,OCC,EVENT,CARDINAL,LOC,WEBSITE,FAC,LAW,TIME,MONEY,CURR,LANGUAGE,PERCENT,PRODUCT,QUANTITY,UNIT
+GPE,2493,22,2,3,3,1,1,2,2,2,0,0,0,0,0,0,0,0,0,0,0
+ORG,20,2504,1,4,3,2,2,4,1,2,0,3,0,0,0,0,0,0,0,0,0
+DATE,2,0,1844,1,1,2,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0
+PERS,6,4,1,902,5,0,5,0,0,0,1,3,0,0,0,0,0,0,0,0,0
+NORP,6,17,0,8,883,0,3,2,1,0,0,1,0,0,0,0,0,0,0,0,0
+ORDINAL,3,2,1,0,3,679,0,0,2,0,0,1,0,0,0,0,0,0,0,0,0
+OCC,1,1,0,2,1,0,719,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+EVENT,3,10,1,0,1,0,1,343,1,0,0,2,0,0,0,0,0,0,0,0,0
+CARDINAL,0,0,1,0,0,4,0,0,318,0,0,0,0,1,0,0,0,0,0,0,0
+LOC,4,1,0,0,1,0,0,0,0,215,0,1,0,0,0,0,0,0,0,0,0
+WEBSITE,0,1,0,0,0,0,0,0,0,0,110,0,0,0,0,0,0,0,0,0,0
+FAC,5,2,0,0,0,0,0,1,0,0,0,109,0,0,0,0,0,0,0,0,0
+LAW,0,0,0,0,0,0,0,0,0,0,0,0,83,0,0,0,0,0,0,0,0
+TIME,1,1,3,0,0,0,0,0,0,1,0,0,0,68,0,0,1,0,0,0,0
+MONEY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,41,0,0,0,0,0,0
+CURR,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,38,0,0,0,0,0
+LANGUAGE,1,2,0,0,1,0,0,0,0,0,0,0,0,0,0,0,26,0,0,0,0
+PERCENT,0,0,0,0,0,0,0,0,2,0,0,0,0,0,1,0,0,26,0,0,0
+PRODUCT,0,1,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,14,0,0
+QUANTITY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,9,1
+UNIT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9

confusion_matrices/validation_unseen_sentence_confusion_matrix.csv

@@ -0,0 +1,22 @@
+,GPE,ORG,DATE,PERS,NORP,ORDINAL,OCC,EVENT,CARDINAL,LOC,WEBSITE,FAC,LAW,TIME,MONEY,CURR,LANGUAGE,PERCENT,PRODUCT,QUANTITY,UNIT
+GPE,149,5,1,1,0,0,0,1,0,1,0,0,0,0,0,0,1,0,0,0,0
+ORG,1,110,0,0,2,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0
+DATE,1,0,122,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+PERS,1,1,0,140,2,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+NORP,1,2,0,3,94,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0
+ORDINAL,0,0,0,2,1,86,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+OCC,1,0,0,3,0,0,40,0,0,0,1,0,0,0,0,0,0,0,0,0,0
+EVENT,1,0,1,0,1,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0,0
+CARDINAL,0,0,0,0,0,2,0,0,47,0,0,0,0,0,0,0,0,0,0,0,0
+LOC,1,1,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0,0,0,0,0
+WEBSITE,0,0,0,0,0,0,0,0,0,0,37,0,0,0,0,0,0,0,0,0,0
+FAC,1,0,0,0,0,0,0,0,0,0,0,12,0,0,0,0,0,0,0,0,0
+LAW,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0,0
+TIME,1,0,1,0,0,2,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0
+MONEY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0
+CURR,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,1,0,0,0,0,0
+LANGUAGE,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,2,0,0,0,0
+PERCENT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0
+PRODUCT,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0
+QUANTITY,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+UNIT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1

dataset_evaluation_category_counts.csv

@@ -0,0 +1,85 @@
+split,type,evaluation_category,count
+validation,GPE,unseen_sentence,159
+validation,GPE,seen_sentence_new_entity,2531
+validation,ORG,unseen_sentence,114
+validation,ORG,seen_sentence_new_entity,2546
+validation,DATE,unseen_sentence,124
+validation,DATE,seen_sentence_new_entity,1852
+validation,PERS,unseen_sentence,145
+validation,PERS,seen_sentence_new_entity,927
+validation,NORP,unseen_sentence,101
+validation,NORP,seen_sentence_new_entity,921
+validation,ORDINAL,unseen_sentence,89
+validation,ORDINAL,seen_sentence_new_entity,691
+validation,OCC,unseen_sentence,45
+validation,OCC,seen_sentence_new_entity,724
+validation,EVENT,unseen_sentence,15
+validation,EVENT,seen_sentence_new_entity,362
+validation,CARDINAL,unseen_sentence,49
+validation,CARDINAL,seen_sentence_new_entity,324
+validation,LOC,unseen_sentence,17
+validation,LOC,seen_sentence_new_entity,222
+validation,WEBSITE,unseen_sentence,37
+validation,WEBSITE,seen_sentence_new_entity,111
+validation,FAC,unseen_sentence,13
+validation,FAC,seen_sentence_new_entity,117
+validation,LAW,unseen_sentence,7
+validation,LAW,seen_sentence_new_entity,83
+validation,TIME,unseen_sentence,12
+validation,TIME,seen_sentence_new_entity,75
+validation,MONEY,unseen_sentence,1
+validation,MONEY,seen_sentence_new_entity,41
+validation,CURR,unseen_sentence,2
+validation,CURR,seen_sentence_new_entity,39
+validation,LANGUAGE,unseen_sentence,3
+validation,LANGUAGE,seen_sentence_new_entity,30
+validation,PERCENT,unseen_sentence,2
+validation,PERCENT,seen_sentence_new_entity,29
+validation,PRODUCT,unseen_sentence,3
+validation,PRODUCT,seen_sentence_new_entity,16
+validation,QUANTITY,unseen_sentence,0
+validation,QUANTITY,seen_sentence_new_entity,11
+validation,UNIT,unseen_sentence,1
+validation,UNIT,seen_sentence_new_entity,9
+test,GPE,unseen_sentence,153
+test,GPE,seen_sentence_new_entity,2538
+test,ORG,unseen_sentence,139
+test,ORG,seen_sentence_new_entity,2520
+test,DATE,unseen_sentence,120
+test,DATE,seen_sentence_new_entity,1857
+test,PERS,unseen_sentence,143
+test,PERS,seen_sentence_new_entity,928
+test,NORP,unseen_sentence,125
+test,NORP,seen_sentence_new_entity,896
+test,ORDINAL,unseen_sentence,82
+test,ORDINAL,seen_sentence_new_entity,699
+test,OCC,unseen_sentence,50
+test,OCC,seen_sentence_new_entity,720
+test,EVENT,unseen_sentence,14
+test,EVENT,seen_sentence_new_entity,363
+test,CARDINAL,unseen_sentence,49
+test,CARDINAL,seen_sentence_new_entity,323
+test,LOC,unseen_sentence,15
+test,LOC,seen_sentence_new_entity,224
+test,WEBSITE,unseen_sentence,52
+test,WEBSITE,seen_sentence_new_entity,96
+test,FAC,unseen_sentence,8
+test,FAC,seen_sentence_new_entity,121
+test,LAW,unseen_sentence,8
+test,LAW,seen_sentence_new_entity,83
+test,TIME,unseen_sentence,19
+test,TIME,seen_sentence_new_entity,68
+test,MONEY,unseen_sentence,1
+test,MONEY,seen_sentence_new_entity,41
+test,CURR,unseen_sentence,3
+test,CURR,seen_sentence_new_entity,38
+test,LANGUAGE,unseen_sentence,8
+test,LANGUAGE,seen_sentence_new_entity,25
+test,PERCENT,unseen_sentence,2
+test,PERCENT,seen_sentence_new_entity,29
+test,PRODUCT,unseen_sentence,2
+test,PRODUCT,seen_sentence_new_entity,17
+test,QUANTITY,unseen_sentence,1
+test,QUANTITY,seen_sentence_new_entity,9
+test,UNIT,unseen_sentence,0
+test,UNIT,seen_sentence_new_entity,11

dataset_split_type_counts.csv

@@ -0,0 +1,22 @@
+type,train,validation,test,total,train_pct,validation_pct,test_pct
+GPE,21521,2690,2691,26902,0.7999776968255148,0.09999256560850495,0.10002973756598023
+ORG,21274,2660,2659,26593,0.7999849584477118,0.10002632271650434,0.09998871883578385
+DATE,15809,1976,1977,19762,0.7999696387005364,0.09998987956684546,0.10004048173261816
+PERS,8569,1072,1071,10712,0.7999439880507841,0.10007468259895444,0.09998132935026138
+NORP,8174,1022,1021,10217,0.8000391504355486,0.10002936282666144,0.09993148673778995
+ORDINAL,6246,780,781,7807,0.8000512360701935,0.09991033687716153,0.10003842705264507
+OCC,6157,769,770,7696,0.8000259875259875,0.09992203742203742,0.10005197505197505
+EVENT,3015,377,377,3769,0.7999469355266648,0.10002653223666755,0.10002653223666755
+CARDINAL,2979,373,372,3724,0.7999462943071965,0.10016111707841031,0.09989258861439312
+LOC,1910,239,239,2388,0.7998324958123953,0.10008375209380234,0.10008375209380234
+WEBSITE,1182,148,148,1478,0.7997293640054127,0.10013531799729364,0.10013531799729364
+FAC,1037,130,129,1296,0.8001543209876543,0.10030864197530864,0.09953703703703703
+LAW,723,90,91,904,0.7997787610619469,0.09955752212389381,0.1006637168141593
+TIME,697,87,87,871,0.8002296211251435,0.09988518943742825,0.09988518943742825
+MONEY,336,42,42,420,0.8,0.1,0.1
+CURR,329,41,41,411,0.8004866180048662,0.09975669099756691,0.09975669099756691
+LANGUAGE,266,33,33,332,0.8012048192771084,0.09939759036144578,0.09939759036144578
+PERCENT,250,31,31,312,0.8012820512820513,0.09935897435897435,0.09935897435897435
+PRODUCT,152,19,19,190,0.8,0.1,0.1
+QUANTITY,85,11,10,106,0.8018867924528302,0.10377358490566038,0.09433962264150944
+UNIT,85,10,11,106,0.8018867924528302,0.09433962264150944,0.10377358490566038

dataset_validation.json

@@ -0,0 +1,790 @@
+{
+  "validated_at_utc": "2026-07-12T15:30:12+00:00",
+  "dataset_repo_id": "U4RASD/TypePrediction",
+  "files": {
+    "train": {
+      "name": "type_predictor_train.jsonl",
+      "path": "/workspace/.cache/huggingface/hub/datasets--U4RASD--TypePrediction/snapshots/2cc13badf71d134057168b7e96cf78d1fdfdc4cc/type_predictor_train.jsonl",
+      "sha256": "20061ac6df35353e01977a85f0d6fc1466a609a093ef76a06a90f7580cbb65f8"
+    },
+    "validation": {
+      "name": "type_predictor_val.jsonl",
+      "path": "/workspace/.cache/huggingface/hub/datasets--U4RASD--TypePrediction/snapshots/2cc13badf71d134057168b7e96cf78d1fdfdc4cc/type_predictor_val.jsonl",
+      "sha256": "099b991973d0ba1f747c3959cbcc8a1eb98e96beadf10d7046451cfc094a19ec"
+    },
+    "test": {
+      "name": "type_predictor_test.jsonl",
+      "path": "/workspace/.cache/huggingface/hub/datasets--U4RASD--TypePrediction/snapshots/2cc13badf71d134057168b7e96cf78d1fdfdc4cc/type_predictor_test.jsonl",
+      "sha256": "7b494a3b7c1f3dcc14aaa9900b18dc983b4c44a8dcd02426068f32bed4f2ca0e"
+    }
+  },
+  "strict_expected_counts": true,
+  "train": {
+    "name": "TRAIN",
+    "rows": 100796,
+    "unique_ids": 100796,
+    "type_counts": {
+      "ORG": 21274,
+      "PERS": 8569,
+      "NORP": 8174,
+      "CARDINAL": 2979,
+      "ORDINAL": 6246,
+      "GPE": 21521,
+      "DATE": 15809,
+      "OCC": 6157,
+      "LOC": 1910,
+      "LANGUAGE": 266,
+      "TIME": 697,
+      "FAC": 1037,
+      "EVENT": 3015,
+      "LAW": 723,
+      "MONEY": 336,
+      "WEBSITE": 1182,
+      "CURR": 329,
+      "UNIT": 85,
+      "PERCENT": 250,
+      "PRODUCT": 152,
+      "QUANTITY": 85
+    },
+    "category_counts": {}
+  },
+  "validation": {
+    "name": "VALIDATION",
+    "rows": 12600,
+    "unique_ids": 12600,
+    "type_counts": {
+      "GPE": 2690,
+      "EVENT": 377,
+      "ORG": 2660,
+      "DATE": 1976,
+      "CARDINAL": 373,
+      "OCC": 769,
+      "ORDINAL": 780,
+      "PERCENT": 31,
+      "PERS": 1072,
+      "NORP": 1022,
+      "TIME": 87,
+      "LOC": 239,
+      "FAC": 130,
+      "WEBSITE": 148,
+      "CURR": 41,
+      "LAW": 90,
+      "QUANTITY": 11,
+      "MONEY": 42,
+      "PRODUCT": 19,
+      "LANGUAGE": 33,
+      "UNIT": 10
+    },
+    "category_counts": {
+      "seen_sentence_new_entity": 11661,
+      "unseen_sentence": 939
+    }
+  },
+  "test": {
+    "name": "TEST",
+    "rows": 12600,
+    "unique_ids": 12600,
+    "type_counts": {
+      "DATE": 1977,
+      "ORG": 2659,
+      "PERS": 1071,
+      "NORP": 1021,
+      "EVENT": 377,
+      "GPE": 2691,
+      "OCC": 770,
+      "FAC": 129,
+      "ORDINAL": 781,
+      "WEBSITE": 148,
+      "TIME": 87,
+      "UNIT": 11,
+      "LOC": 239,
+      "MONEY": 42,
+      "CARDINAL": 372,
+      "CURR": 41,
+      "PERCENT": 31,
+      "LAW": 91,
+      "PRODUCT": 19,
+      "LANGUAGE": 33,
+      "QUANTITY": 10
+    },
+    "category_counts": {
+      "seen_sentence_new_entity": 11606,
+      "unseen_sentence": 994
+    }
+  },
+  "all_ids_unique_across_splits": true,
+  "label_list": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "source_summary": {
+    "total_rows": 125996,
+    "unique_sentences": 33933,
+    "raw_mentions_before_dedupe": 408035,
+    "deduped_rows_before_ambiguity_filter": 132763,
+    "duplicates_merged": 275272,
+    "ambiguous_span_groups_excluded": 3383,
+    "ambiguous_candidate_rows_excluded": 6767,
+    "source_breakdown": [
+      {
+        "dataset": "Wojood1_1_flat",
+        "split": "test",
+        "sentences": 6606,
+        "mentions": 14573
+      },
+      {
+        "dataset": "Wojood1_1_flat",
+        "split": "train",
+        "sentences": 23125,
+        "mentions": 49959
+      },
+      {
+        "dataset": "Wojood1_1_flat",
+        "split": "val",
+        "sentences": 3304,
+        "mentions": 7143
+      },
+      {
+        "dataset": "Wojood1_1_nested",
+        "split": "test",
+        "sentences": 6606,
+        "mentions": 18045
+      },
+      {
+        "dataset": "Wojood1_1_nested",
+        "split": "train",
+        "sentences": 23125,
+        "mentions": 62377
+      },
+      {
+        "dataset": "Wojood1_1_nested",
+        "split": "val",
+        "sentences": 3304,
+        "mentions": 8944
+      },
+      {
+        "dataset": "WojoodFine",
+        "split": "test",
+        "sentences": 5748,
+        "mentions": 27848
+      },
+      {
+        "dataset": "WojoodFine",
+        "split": "train",
+        "sentences": 19484,
+        "mentions": 96188
+      },
+      {
+        "dataset": "WojoodFine",
+        "split": "val",
+        "sentences": 2828,
+        "mentions": 13800
+      },
+      {
+        "dataset": "WojoodFine-Flat",
+        "split": "split10",
+        "sentences": 3304,
+        "mentions": 10859
+      },
+      {
+        "dataset": "WojoodFine-Flat",
+        "split": "split20",
+        "sentences": 6606,
+        "mentions": 22208
+      },
+      {
+        "dataset": "WojoodFine-Flat",
+        "split": "split70",
+        "sentences": 23125,
+        "mentions": 76091
+      }
+    ],
+    "type_breakdown": [
+      {
+        "type": "GPE",
+        "count": 26902
+      },
+      {
+        "type": "ORG",
+        "count": 26593
+      },
+      {
+        "type": "DATE",
+        "count": 19762
+      },
+      {
+        "type": "PERS",
+        "count": 10712
+      },
+      {
+        "type": "NORP",
+        "count": 10217
+      },
+      {
+        "type": "ORDINAL",
+        "count": 7807
+      },
+      {
+        "type": "OCC",
+        "count": 7696
+      },
+      {
+        "type": "EVENT",
+        "count": 3769
+      },
+      {
+        "type": "CARDINAL",
+        "count": 3724
+      },
+      {
+        "type": "LOC",
+        "count": 2388
+      },
+      {
+        "type": "WEBSITE",
+        "count": 1478
+      },
+      {
+        "type": "FAC",
+        "count": 1296
+      },
+      {
+        "type": "LAW",
+        "count": 904
+      },
+      {
+        "type": "TIME",
+        "count": 871
+      },
+      {
+        "type": "MONEY",
+        "count": 420
+      },
+      {
+        "type": "CURR",
+        "count": 411
+      },
+      {
+        "type": "LANGUAGE",
+        "count": 332
+      },
+      {
+        "type": "PERCENT",
+        "count": 312
+      },
+      {
+        "type": "PRODUCT",
+        "count": 190
+      },
+      {
+        "type": "QUANTITY",
+        "count": 106
+      },
+      {
+        "type": "UNIT",
+        "count": 106
+      }
+    ]
+  },
+  "source_split_summary": {
+    "random_seed": 42,
+    "requested_split_ratios": {
+      "train": 0.8,
+      "validation": 0.1,
+      "test": 0.1
+    },
+    "totals": {
+      "original_rows": 125996,
+      "train_rows": 100796,
+      "validation_rows": 12600,
+      "test_rows": 12600
+    },
+    "splits": {
+      "train": {
+        "name": "train",
+        "rows": 100796,
+        "percentage": 0.7999936505920823,
+        "unique_sentences": 32409,
+        "type_counts": {
+          "CARDINAL": 2979,
+          "CURR": 329,
+          "DATE": 15809,
+          "EVENT": 3015,
+          "FAC": 1037,
+          "GPE": 21521,
+          "LANGUAGE": 266,
+          "LAW": 723,
+          "LOC": 1910,
+          "MONEY": 336,
+          "NORP": 8174,
+          "OCC": 6157,
+          "ORDINAL": 6246,
+          "ORG": 21274,
+          "PERCENT": 250,
+          "PERS": 8569,
+          "PRODUCT": 152,
+          "QUANTITY": 85,
+          "TIME": 697,
+          "UNIT": 85,
+          "WEBSITE": 1182
+        }
+      },
+      "validation": {
+        "name": "validation",
+        "rows": 12600,
+        "percentage": 0.10000317470395885,
+        "unique_sentences": 10139,
+        "type_counts": {
+          "CARDINAL": 373,
+          "CURR": 41,
+          "DATE": 1976,
+          "EVENT": 377,
+          "FAC": 130,
+          "GPE": 2690,
+          "LANGUAGE": 33,
+          "LAW": 90,
+          "LOC": 239,
+          "MONEY": 42,
+          "NORP": 1022,
+          "OCC": 769,
+          "ORDINAL": 780,
+          "ORG": 2660,
+          "PERCENT": 31,
+          "PERS": 1072,
+          "PRODUCT": 19,
+          "QUANTITY": 11,
+          "TIME": 87,
+          "UNIT": 10,
+          "WEBSITE": 148
+        }
+      },
+      "test": {
+        "name": "test",
+        "rows": 12600,
+        "percentage": 0.10000317470395885,
+        "unique_sentences": 10161,
+        "type_counts": {
+          "CARDINAL": 372,
+          "CURR": 41,
+          "DATE": 1977,
+          "EVENT": 377,
+          "FAC": 129,
+          "GPE": 2691,
+          "LANGUAGE": 33,
+          "LAW": 91,
+          "LOC": 239,
+          "MONEY": 42,
+          "NORP": 1021,
+          "OCC": 770,
+          "ORDINAL": 781,
+          "ORG": 2659,
+          "PERCENT": 31,
+          "PERS": 1071,
+          "PRODUCT": 19,
+          "QUANTITY": 10,
+          "TIME": 87,
+          "UNIT": 11,
+          "WEBSITE": 148
+        }
+      }
+    },
+    "per_type_counts": {
+      "CARDINAL": {
+        "original": 3724,
+        "train": 2979,
+        "validation": 373,
+        "test": 372,
+        "train_ratio": 0.7999462943071965,
+        "validation_ratio": 0.10016111707841031,
+        "test_ratio": 0.09989258861439312
+      },
+      "CURR": {
+        "original": 411,
+        "train": 329,
+        "validation": 41,
+        "test": 41,
+        "train_ratio": 0.8004866180048662,
+        "validation_ratio": 0.09975669099756691,
+        "test_ratio": 0.09975669099756691
+      },
+      "DATE": {
+        "original": 19762,
+        "train": 15809,
+        "validation": 1976,
+        "test": 1977,
+        "train_ratio": 0.7999696387005364,
+        "validation_ratio": 0.09998987956684546,
+        "test_ratio": 0.10004048173261816
+      },
+      "EVENT": {
+        "original": 3769,
+        "train": 3015,
+        "validation": 377,
+        "test": 377,
+        "train_ratio": 0.7999469355266648,
+        "validation_ratio": 0.10002653223666755,
+        "test_ratio": 0.10002653223666755
+      },
+      "FAC": {
+        "original": 1296,
+        "train": 1037,
+        "validation": 130,
+        "test": 129,
+        "train_ratio": 0.8001543209876543,
+        "validation_ratio": 0.10030864197530864,
+        "test_ratio": 0.09953703703703703
+      },
+      "GPE": {
+        "original": 26902,
+        "train": 21521,
+        "validation": 2690,
+        "test": 2691,
+        "train_ratio": 0.7999776968255148,
+        "validation_ratio": 0.09999256560850495,
+        "test_ratio": 0.10002973756598023
+      },
+      "LANGUAGE": {
+        "original": 332,
+        "train": 266,
+        "validation": 33,
+        "test": 33,
+        "train_ratio": 0.8012048192771084,
+        "validation_ratio": 0.09939759036144578,
+        "test_ratio": 0.09939759036144578
+      },
+      "LAW": {
+        "original": 904,
+        "train": 723,
+        "validation": 90,
+        "test": 91,
+        "train_ratio": 0.7997787610619469,
+        "validation_ratio": 0.09955752212389381,
+        "test_ratio": 0.1006637168141593
+      },
+      "LOC": {
+        "original": 2388,
+        "train": 1910,
+        "validation": 239,
+        "test": 239,
+        "train_ratio": 0.7998324958123953,
+        "validation_ratio": 0.10008375209380234,
+        "test_ratio": 0.10008375209380234
+      },
+      "MONEY": {
+        "original": 420,
+        "train": 336,
+        "validation": 42,
+        "test": 42,
+        "train_ratio": 0.8,
+        "validation_ratio": 0.1,
+        "test_ratio": 0.1
+      },
+      "NORP": {
+        "original": 10217,
+        "train": 8174,
+        "validation": 1022,
+        "test": 1021,
+        "train_ratio": 0.8000391504355486,
+        "validation_ratio": 0.10002936282666144,
+        "test_ratio": 0.09993148673778995
+      },
+      "OCC": {
+        "original": 7696,
+        "train": 6157,
+        "validation": 769,
+        "test": 770,
+        "train_ratio": 0.8000259875259875,
+        "validation_ratio": 0.09992203742203742,
+        "test_ratio": 0.10005197505197505
+      },
+      "ORDINAL": {
+        "original": 7807,
+        "train": 6246,
+        "validation": 780,
+        "test": 781,
+        "train_ratio": 0.8000512360701935,
+        "validation_ratio": 0.09991033687716153,
+        "test_ratio": 0.10003842705264507
+      },
+      "ORG": {
+        "original": 26593,
+        "train": 21274,
+        "validation": 2660,
+        "test": 2659,
+        "train_ratio": 0.7999849584477118,
+        "validation_ratio": 0.10002632271650434,
+        "test_ratio": 0.09998871883578385
+      },
+      "PERCENT": {
+        "original": 312,
+        "train": 250,
+        "validation": 31,
+        "test": 31,
+        "train_ratio": 0.8012820512820513,
+        "validation_ratio": 0.09935897435897435,
+        "test_ratio": 0.09935897435897435
+      },
+      "PERS": {
+        "original": 10712,
+        "train": 8569,
+        "validation": 1072,
+        "test": 1071,
+        "train_ratio": 0.7999439880507841,
+        "validation_ratio": 0.10007468259895444,
+        "test_ratio": 0.09998132935026138
+      },
+      "PRODUCT": {
+        "original": 190,
+        "train": 152,
+        "validation": 19,
+        "test": 19,
+        "train_ratio": 0.8,
+        "validation_ratio": 0.1,
+        "test_ratio": 0.1
+      },
+      "QUANTITY": {
+        "original": 106,
+        "train": 85,
+        "validation": 11,
+        "test": 10,
+        "train_ratio": 0.8018867924528302,
+        "validation_ratio": 0.10377358490566038,
+        "test_ratio": 0.09433962264150944
+      },
+      "TIME": {
+        "original": 871,
+        "train": 697,
+        "validation": 87,
+        "test": 87,
+        "train_ratio": 0.8002296211251435,
+        "validation_ratio": 0.09988518943742825,
+        "test_ratio": 0.09988518943742825
+      },
+      "UNIT": {
+        "original": 106,
+        "train": 85,
+        "validation": 10,
+        "test": 11,
+        "train_ratio": 0.8018867924528302,
+        "validation_ratio": 0.09433962264150944,
+        "test_ratio": 0.10377358490566038
+      },
+      "WEBSITE": {
+        "original": 1478,
+        "train": 1182,
+        "validation": 148,
+        "test": 148,
+        "train_ratio": 0.7997293640054127,
+        "validation_ratio": 0.10013531799729364,
+        "test_ratio": 0.10013531799729364
+      }
+    },
+    "evaluation_category_counts": {
+      "validation": {
+        "seen_sentence_new_entity": 11661,
+        "unseen_sentence": 939
+      },
+      "test": {
+        "seen_sentence_new_entity": 11606,
+        "unseen_sentence": 994
+      }
+    },
+    "evaluation_category_counts_by_type": {
+      "validation": {
+        "CARDINAL": {
+          "unseen_sentence": 49,
+          "seen_sentence_new_entity": 324
+        },
+        "CURR": {
+          "unseen_sentence": 2,
+          "seen_sentence_new_entity": 39
+        },
+        "DATE": {
+          "unseen_sentence": 124,
+          "seen_sentence_new_entity": 1852
+        },
+        "EVENT": {
+          "unseen_sentence": 15,
+          "seen_sentence_new_entity": 362
+        },
+        "FAC": {
+          "unseen_sentence": 13,
+          "seen_sentence_new_entity": 117
+        },
+        "GPE": {
+          "unseen_sentence": 159,
+          "seen_sentence_new_entity": 2531
+        },
+        "LANGUAGE": {
+          "unseen_sentence": 3,
+          "seen_sentence_new_entity": 30
+        },
+        "LAW": {
+          "unseen_sentence": 7,
+          "seen_sentence_new_entity": 83
+        },
+        "LOC": {
+          "unseen_sentence": 17,
+          "seen_sentence_new_entity": 222
+        },
+        "MONEY": {
+          "unseen_sentence": 1,
+          "seen_sentence_new_entity": 41
+        },
+        "NORP": {
+          "unseen_sentence": 101,
+          "seen_sentence_new_entity": 921
+        },
+        "OCC": {
+          "unseen_sentence": 45,
+          "seen_sentence_new_entity": 724
+        },
+        "ORDINAL": {
+          "unseen_sentence": 89,
+          "seen_sentence_new_entity": 691
+        },
+        "ORG": {
+          "unseen_sentence": 114,
+          "seen_sentence_new_entity": 2546
+        },
+        "PERCENT": {
+          "unseen_sentence": 2,
+          "seen_sentence_new_entity": 29
+        },
+        "PERS": {
+          "unseen_sentence": 145,
+          "seen_sentence_new_entity": 927
+        },
+        "PRODUCT": {
+          "unseen_sentence": 3,
+          "seen_sentence_new_entity": 16
+        },
+        "QUANTITY": {
+          "unseen_sentence": 0,
+          "seen_sentence_new_entity": 11
+        },
+        "TIME": {
+          "unseen_sentence": 12,
+          "seen_sentence_new_entity": 75
+        },
+        "UNIT": {
+          "unseen_sentence": 1,
+          "seen_sentence_new_entity": 9
+        },
+        "WEBSITE": {
+          "unseen_sentence": 37,
+          "seen_sentence_new_entity": 111
+        }
+      },
+      "test": {
+        "CARDINAL": {
+          "unseen_sentence": 49,
+          "seen_sentence_new_entity": 323
+        },
+        "CURR": {
+          "unseen_sentence": 3,
+          "seen_sentence_new_entity": 38
+        },
+        "DATE": {
+          "unseen_sentence": 120,
+          "seen_sentence_new_entity": 1857
+        },
+        "EVENT": {
+          "unseen_sentence": 14,
+          "seen_sentence_new_entity": 363
+        },
+        "FAC": {
+          "unseen_sentence": 8,
+          "seen_sentence_new_entity": 121
+        },
+        "GPE": {
+          "unseen_sentence": 153,
+          "seen_sentence_new_entity": 2538
+        },
+        "LANGUAGE": {
+          "unseen_sentence": 8,
+          "seen_sentence_new_entity": 25
+        },
+        "LAW": {
+          "unseen_sentence": 8,
+          "seen_sentence_new_entity": 83
+        },
+        "LOC": {
+          "unseen_sentence": 15,
+          "seen_sentence_new_entity": 224
+        },
+        "MONEY": {
+          "unseen_sentence": 1,
+          "seen_sentence_new_entity": 41
+        },
+        "NORP": {
+          "unseen_sentence": 125,
+          "seen_sentence_new_entity": 896
+        },
+        "OCC": {
+          "unseen_sentence": 50,
+          "seen_sentence_new_entity": 720
+        },
+        "ORDINAL": {
+          "unseen_sentence": 82,
+          "seen_sentence_new_entity": 699
+        },
+        "ORG": {
+          "unseen_sentence": 139,
+          "seen_sentence_new_entity": 2520
+        },
+        "PERCENT": {
+          "unseen_sentence": 2,
+          "seen_sentence_new_entity": 29
+        },
+        "PERS": {
+          "unseen_sentence": 143,
+          "seen_sentence_new_entity": 928
+        },
+        "PRODUCT": {
+          "unseen_sentence": 2,
+          "seen_sentence_new_entity": 17
+        },
+        "QUANTITY": {
+          "unseen_sentence": 1,
+          "seen_sentence_new_entity": 9
+        },
+        "TIME": {
+          "unseen_sentence": 19,
+          "seen_sentence_new_entity": 68
+        },
+        "UNIT": {
+          "unseen_sentence": 0,
+          "seen_sentence_new_entity": 11
+        },
+        "WEBSITE": {
+          "unseen_sentence": 52,
+          "seen_sentence_new_entity": 96
+        }
+      }
+    },
+    "validation": {
+      "ok": true,
+      "errors": []
+    }
+  }
+}

encoding_summary.json

@@ -0,0 +1,61 @@
+{
+  "max_length": 512,
+  "context_candidates": [
+    null,
+    500,
+    300,
+    150,
+    80,
+    30,
+    0
+  ],
+  "marker_tokens": [
+    "[ENT]",
+    "[/ENT]"
+  ],
+  "marker_token_ids": {
+    "[ENT]": 65000,
+    "[/ENT]": 65001
+  },
+  "splits": {
+    "train": {
+      "rows": 100796,
+      "token_length_min": 5,
+      "token_length_mean": 53.74587285209731,
+      "token_length_median": 31.0,
+      "token_length_p95": 174.0,
+      "token_length_p99": 393.25000000001455,
+      "token_length_max": 512,
+      "context_counts": {
+        "None": 100590,
+        "500": 206
+      }
+    },
+    "validation": {
+      "rows": 12600,
+      "token_length_min": 5,
+      "token_length_mean": 52.443174603174604,
+      "token_length_median": 31.0,
+      "token_length_p95": 165.0,
+      "token_length_p99": 393.0,
+      "token_length_max": 512,
+      "context_counts": {
+        "None": 12570,
+        "500": 30
+      }
+    },
+    "test": {
+      "rows": 12600,
+      "token_length_min": 5,
+      "token_length_mean": 53.945238095238096,
+      "token_length_median": 31.0,
+      "token_length_p95": 178.0,
+      "token_length_p99": 409.0,
+      "token_length_max": 512,
+      "context_counts": {
+        "None": 12575,
+        "500": 25
+      }
+    }
+  }
+}

evaluation_results.json

@@ -0,0 +1,358 @@
+{
+  "status": "completed",
+  "completed_at_utc": "2026-07-12T18:56:34+00:00",
+  "dataset_repo_id": "U4RASD/TypePrediction",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "architecture": {
+    "architecture": "NeoAraBERTTypePredictor",
+    "task": "mention-level 21-class Arabic entity typing",
+    "base_model": "U4RASD/NeoAraBERT",
+    "input_markers": [
+      "[ENT]",
+      "[/ENT]"
+    ],
+    "representation": "last hidden state at token position 0 (CLS)",
+    "dropout": 0.1,
+    "classifier": [
+      768,
+      21
+    ],
+    "loss": "ordinary unweighted multiclass cross-entropy",
+    "num_labels": 21,
+    "labels": [
+      "GPE",
+      "ORG",
+      "DATE",
+      "PERS",
+      "NORP",
+      "ORDINAL",
+      "OCC",
+      "EVENT",
+      "CARDINAL",
+      "LOC",
+      "WEBSITE",
+      "FAC",
+      "LAW",
+      "TIME",
+      "MONEY",
+      "CURR",
+      "LANGUAGE",
+      "PERCENT",
+      "PRODUCT",
+      "QUANTITY",
+      "UNIT"
+    ],
+    "tokenizer_size": 65002,
+    "hidden_size": 768,
+    "max_length": 512,
+    "context_candidates": [
+      null,
+      500,
+      300,
+      150,
+      80,
+      30,
+      0
+    ]
+  },
+  "dataset_counts": {
+    "train": 100796,
+    "validation": 12600,
+    "test": 12600
+  },
+  "validation_results": {
+    "overall": {
+      "subset": "overall",
+      "rows": 12600,
+      "types_present": [
+        "GPE",
+        "ORG",
+        "DATE",
+        "PERS",
+        "NORP",
+        "ORDINAL",
+        "OCC",
+        "EVENT",
+        "CARDINAL",
+        "LOC",
+        "WEBSITE",
+        "FAC",
+        "LAW",
+        "TIME",
+        "MONEY",
+        "CURR",
+        "LANGUAGE",
+        "PERCENT",
+        "PRODUCT",
+        "QUANTITY",
+        "UNIT"
+      ],
+      "num_types_present": 21,
+      "accuracy": 0.9778571428571429,
+      "micro_precision": 0.9778571428571429,
+      "micro_recall": 0.9778571428571429,
+      "micro_f1": 0.9778571428571429,
+      "macro_precision_present_types": 0.9725024753902682,
+      "macro_recall_present_types": 0.9475659044801065,
+      "macro_f1_present_types": 0.9587281193044483,
+      "macro_precision_all_21": 0.9725024753902682,
+      "macro_recall_all_21": 0.9475659044801065,
+      "macro_f1_all_21": 0.9587281193044483,
+      "weighted_precision": 0.977879248064854,
+      "weighted_recall": 0.9778571428571429,
+      "weighted_f1": 0.9777906938936711,
+      "split": "validation",
+      "checkpoint": "best",
+      "model_repo_id": "U4RASD/TypePredictor",
+      "generated_at_utc": "2026-07-12T18:55:47+00:00"
+    },
+    "unseen_sentence": {
+      "subset": "unseen_sentence",
+      "rows": 939,
+      "types_present": [
+        "GPE",
+        "ORG",
+        "DATE",
+        "PERS",
+        "NORP",
+        "ORDINAL",
+        "OCC",
+        "EVENT",
+        "CARDINAL",
+        "LOC",
+        "WEBSITE",
+        "FAC",
+        "LAW",
+        "TIME",
+        "MONEY",
+        "CURR",
+        "LANGUAGE",
+        "PERCENT",
+        "PRODUCT",
+        "UNIT"
+      ],
+      "num_types_present": 20,
+      "accuracy": 0.9456869009584664,
+      "micro_precision": 0.9456869009584664,
+      "micro_recall": 0.9456869009584664,
+      "micro_f1": 0.9456869009584664,
+      "macro_precision_present_types": 0.9465091355300324,
+      "macro_recall_present_types": 0.8850947519406029,
+      "macro_f1_present_types": 0.9088413774156396,
+      "macro_precision_all_21": 0.9014372719333641,
+      "macro_recall_all_21": 0.8429473828005742,
+      "macro_f1_all_21": 0.8655632165863235,
+      "weighted_precision": 0.9458822212425712,
+      "weighted_recall": 0.9456869009584664,
+      "weighted_f1": 0.945010776949412,
+      "split": "validation",
+      "checkpoint": "best",
+      "model_repo_id": "U4RASD/TypePredictor",
+      "generated_at_utc": "2026-07-12T18:55:47+00:00"
+    },
+    "seen_sentence_new_entity": {
+      "subset": "seen_sentence_new_entity",
+      "rows": 11661,
+      "types_present": [
+        "GPE",
+        "ORG",
+        "DATE",
+        "PERS",
+        "NORP",
+        "ORDINAL",
+        "OCC",
+        "EVENT",
+        "CARDINAL",
+        "LOC",
+        "WEBSITE",
+        "FAC",
+        "LAW",
+        "TIME",
+        "MONEY",
+        "CURR",
+        "LANGUAGE",
+        "PERCENT",
+        "PRODUCT",
+        "QUANTITY",
+        "UNIT"
+      ],
+      "num_types_present": 21,
+      "accuracy": 0.9804476459994854,
+      "micro_precision": 0.9804476459994854,
+      "micro_recall": 0.9804476459994854,
+      "micro_f1": 0.9804476459994854,
+      "macro_precision_present_types": 0.9751134735007793,
+      "macro_recall_present_types": 0.9537701979295099,
+      "macro_f1_present_types": 0.963262123278673,
+      "macro_precision_all_21": 0.9751134735007793,
+      "macro_recall_all_21": 0.9537701979295099,
+      "macro_f1_all_21": 0.963262123278673,
+      "weighted_precision": 0.980497781530163,
+      "weighted_recall": 0.9804476459994854,
+      "weighted_f1": 0.9804033222521132,
+      "split": "validation",
+      "checkpoint": "best",
+      "model_repo_id": "U4RASD/TypePredictor",
+      "generated_at_utc": "2026-07-12T18:55:47+00:00"
+    }
+  },
+  "test_results": {
+    "overall": {
+      "subset": "overall",
+      "rows": 12600,
+      "types_present": [
+        "GPE",
+        "ORG",
+        "DATE",
+        "PERS",
+        "NORP",
+        "ORDINAL",
+        "OCC",
+        "EVENT",
+        "CARDINAL",
+        "LOC",
+        "WEBSITE",
+        "FAC",
+        "LAW",
+        "TIME",
+        "MONEY",
+        "CURR",
+        "LANGUAGE",
+        "PERCENT",
+        "PRODUCT",
+        "QUANTITY",
+        "UNIT"
+      ],
+      "num_types_present": 21,
+      "accuracy": 0.9793650793650793,
+      "micro_precision": 0.9793650793650793,
+      "micro_recall": 0.9793650793650793,
+      "micro_f1": 0.9793650793650793,
+      "macro_precision_present_types": 0.9672581856551346,
+      "macro_recall_present_types": 0.9534679938651157,
+      "macro_f1_present_types": 0.9599200012897606,
+      "macro_precision_all_21": 0.9672581856551346,
+      "macro_recall_all_21": 0.9534679938651157,
+      "macro_f1_all_21": 0.9599200012897606,
+      "weighted_precision": 0.9793254669511197,
+      "weighted_recall": 0.9793650793650793,
+      "weighted_f1": 0.9792974524068463,
+      "split": "test",
+      "checkpoint": "best",
+      "model_repo_id": "U4RASD/TypePredictor",
+      "generated_at_utc": "2026-07-12T18:56:34+00:00"
+    },
+    "unseen_sentence": {
+      "subset": "unseen_sentence",
+      "rows": 994,
+      "types_present": [
+        "GPE",
+        "ORG",
+        "DATE",
+        "PERS",
+        "NORP",
+        "ORDINAL",
+        "OCC",
+        "EVENT",
+        "CARDINAL",
+        "LOC",
+        "WEBSITE",
+        "FAC",
+        "LAW",
+        "TIME",
+        "MONEY",
+        "CURR",
+        "LANGUAGE",
+        "PERCENT",
+        "PRODUCT",
+        "QUANTITY"
+      ],
+      "num_types_present": 20,
+      "accuracy": 0.9627766599597586,
+      "micro_precision": 0.9627766599597586,
+      "micro_recall": 0.9627766599597586,
+      "micro_f1": 0.9627766599597586,
+      "macro_precision_present_types": 0.9052965403310622,
+      "macro_recall_present_types": 0.886274373082116,
+      "macro_f1_present_types": 0.8931735856196752,
+      "macro_precision_all_21": 0.8621871812676783,
+      "macro_recall_all_21": 0.8440708315067772,
+      "macro_f1_all_21": 0.8506415101139764,
+      "weighted_precision": 0.9615439764829642,
+      "weighted_recall": 0.9627766599597586,
+      "weighted_f1": 0.9617803672434098,
+      "split": "test",
+      "checkpoint": "best",
+      "model_repo_id": "U4RASD/TypePredictor",
+      "generated_at_utc": "2026-07-12T18:56:34+00:00"
+    },
+    "seen_sentence_new_entity": {
+      "subset": "seen_sentence_new_entity",
+      "rows": 11606,
+      "types_present": [
+        "GPE",
+        "ORG",
+        "DATE",
+        "PERS",
+        "NORP",
+        "ORDINAL",
+        "OCC",
+        "EVENT",
+        "CARDINAL",
+        "LOC",
+        "WEBSITE",
+        "FAC",
+        "LAW",
+        "TIME",
+        "MONEY",
+        "CURR",
+        "LANGUAGE",
+        "PERCENT",
+        "PRODUCT",
+        "QUANTITY",
+        "UNIT"
+      ],
+      "num_types_present": 21,
+      "accuracy": 0.9807858004480441,
+      "micro_precision": 0.9807858004480441,
+      "micro_recall": 0.9807858004480441,
+      "micro_f1": 0.9807858004480441,
+      "macro_precision_present_types": 0.970282742754196,
+      "macro_recall_present_types": 0.9589517514121707,
+      "macro_f1_present_types": 0.9641131075953279,
+      "macro_precision_all_21": 0.970282742754196,
+      "macro_recall_all_21": 0.9589517514121707,
+      "macro_f1_all_21": 0.9641131075953279,
+      "weighted_precision": 0.9807493542827415,
+      "weighted_recall": 0.9807858004480441,
+      "weighted_f1": 0.9807123412635281,
+      "split": "test",
+      "checkpoint": "best",
+      "model_repo_id": "U4RASD/TypePredictor",
+      "generated_at_utc": "2026-07-12T18:56:34+00:00"
+    }
+  },
+  "checkpoint_summary": {
+    "best_checkpoint_source": "outputs/TypePredictor/checkpoints/checkpoint-25000",
+    "best_checkpoint_preserved": "outputs/TypePredictor/best_checkpoint",
+    "best_step": 25000,
+    "best_epoch": 3.9682923925552602,
+    "best_validation_macro_f1": 0.9587281193044483,
+    "latest_checkpoint_source": "outputs/TypePredictor/checkpoints/checkpoint-25196",
+    "latest_checkpoint_preserved": "outputs/TypePredictor/latest_checkpoint",
+    "latest_step": 25196,
+    "latest_epoch": 3.9994047382832654,
+    "best_and_latest_same_source": false,
+    "root_released_model": "best checkpoint"
+  },
+  "training_wall_seconds": 12262.467122793198,
+  "train_metrics": {
+    "train_runtime": 12261.7153,
+    "train_samples_per_second": 32.882,
+    "train_steps_per_second": 2.055,
+    "total_flos": 0.0,
+    "train_loss": 0.19069333035238045,
+    "epoch": 3.9994047382832654
+  },
+  "trainer_log_history_entries": 554
+}

id2label.json

@@ -0,0 +1,23 @@
+{
+  "0": "GPE",
+  "1": "ORG",
+  "2": "DATE",
+  "3": "PERS",
+  "4": "NORP",
+  "5": "ORDINAL",
+  "6": "OCC",
+  "7": "EVENT",
+  "8": "CARDINAL",
+  "9": "LOC",
+  "10": "WEBSITE",
+  "11": "FAC",
+  "12": "LAW",
+  "13": "TIME",
+  "14": "MONEY",
+  "15": "CURR",
+  "16": "LANGUAGE",
+  "17": "PERCENT",
+  "18": "PRODUCT",
+  "19": "QUANTITY",
+  "20": "UNIT"
+}

inference.py

@@ -0,0 +1,75 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""Safe inference example for U4RASD/TypePredictor."""
+
+import argparse
+import json
+import torch
+from modeling_type_predictor import NeoAraBERTTypePredictor
+
+OPEN_MARKER = "[ENT]"
+CLOSE_MARKER = "[/ENT]"
+
+
+def mark_entity(sentence: str, start_char: int, end_char: int) -> str:
+    entity = sentence[start_char:end_char]
+    return sentence[:start_char] + " [ENT] " + entity + " [/ENT] " + sentence[end_char:]
+
+
+def windowed_marked_text(sentence: str, start_char: int, end_char: int, context_chars):
+    if context_chars is None:
+        return mark_entity(sentence, start_char, end_char)
+    left = max(0, start_char - int(context_chars))
+    right = min(len(sentence), end_char + int(context_chars))
+    window = sentence[left:right]
+    return mark_entity(window, start_char - left, end_char - left)
+
+
+def encode_safely(tokenizer, config, sentence: str, start_char: int, end_char: int):
+    if not (0 <= start_char < end_char <= len(sentence)):
+        raise ValueError("Invalid character span.")
+    open_id = tokenizer.convert_tokens_to_ids(OPEN_MARKER)
+    close_id = tokenizer.convert_tokens_to_ids(CLOSE_MARKER)
+    candidates = config.get("context_candidates", [None, 500, 300, 150, 80, 30, 0])
+    for context in candidates:
+        text = windowed_marked_text(sentence, start_char, end_char, context)
+        batch = tokenizer(
+            text,
+            return_tensors="pt",
+            truncation=True,
+            max_length=int(config.get("max_length", 512)),
+        )
+        ids = batch["input_ids"][0].tolist()
+        if ids.count(open_id) == 1 and ids.count(close_id) == 1 and ids.index(open_id) < ids.index(close_id):
+            return batch, text, context
+    raise RuntimeError("Both entity markers could not be preserved after entity-centered truncation.")
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="U4RASD/TypePredictor")
+    parser.add_argument("--sentence", required=True)
+    parser.add_argument("--start-char", type=int, required=True)
+    parser.add_argument("--end-char", type=int, required=True)
+    args = parser.parse_args()
+
+    model, tokenizer, config = NeoAraBERTTypePredictor.from_pretrained(args.model)
+    model.eval()
+    batch, marked_text, context = encode_safely(
+        tokenizer, config, args.sentence, args.start_char, args.end_char
+    )
+    with torch.no_grad():
+        logits = model(**batch)["logits"]
+        probabilities = torch.softmax(logits, dim=-1)[0]
+    index = int(probabilities.argmax())
+    print(json.dumps({
+        "entity": args.sentence[args.start_char:args.end_char],
+        "predicted_type": config["labels"][index],
+        "confidence": float(probabilities[index]),
+        "context_chars_used": context,
+        "marked_text": marked_text,
+    }, ensure_ascii=False, indent=2))
+
+
+if __name__ == "__main__":
+    main()

label2id.json

@@ -0,0 +1,23 @@
+{
+  "GPE": 0,
+  "ORG": 1,
+  "DATE": 2,
+  "PERS": 3,
+  "NORP": 4,
+  "ORDINAL": 5,
+  "OCC": 6,
+  "EVENT": 7,
+  "CARDINAL": 8,
+  "LOC": 9,
+  "WEBSITE": 10,
+  "FAC": 11,
+  "LAW": 12,
+  "TIME": 13,
+  "MONEY": 14,
+  "CURR": 15,
+  "LANGUAGE": 16,
+  "PERCENT": 17,
+  "PRODUCT": 18,
+  "QUANTITY": 19,
+  "UNIT": 20
+}

metrics/test_overall.json

@@ -0,0 +1,45 @@
+{
+  "subset": "overall",
+  "rows": 12600,
+  "types_present": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "num_types_present": 21,
+  "accuracy": 0.9793650793650793,
+  "micro_precision": 0.9793650793650793,
+  "micro_recall": 0.9793650793650793,
+  "micro_f1": 0.9793650793650793,
+  "macro_precision_present_types": 0.9672581856551346,
+  "macro_recall_present_types": 0.9534679938651157,
+  "macro_f1_present_types": 0.9599200012897606,
+  "macro_precision_all_21": 0.9672581856551346,
+  "macro_recall_all_21": 0.9534679938651157,
+  "macro_f1_all_21": 0.9599200012897606,
+  "weighted_precision": 0.9793254669511197,
+  "weighted_recall": 0.9793650793650793,
+  "weighted_f1": 0.9792974524068463,
+  "split": "test",
+  "checkpoint": "best",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "generated_at_utc": "2026-07-12T18:56:34+00:00"
+}

metrics/test_overall_per_class.csv

@@ -0,0 +1,25 @@
+,precision,recall,f1-score,support
+GPE,0.9738970588235294,0.9843924191750278,0.9791166143041952,2691.0
+ORG,0.9807692307692307,0.9781872884543061,0.9794765580869893,2659.0
+DATE,0.9904378459989934,0.9954476479514416,0.992936427850656,1977.0
+PERS,0.9795348837209302,0.9831932773109243,0.9813606710158435,1071.0
+NORP,0.9765395894428153,0.9784524975514202,0.9774951076320939,1021.0
+ORDINAL,0.9820971867007673,0.9833546734955185,0.982725527831094,781.0
+OCC,0.9870466321243523,0.9896103896103896,0.9883268482490273,770.0
+EVENT,0.9864130434782609,0.9628647214854111,0.974496644295302,377.0
+CARDINAL,0.9779005524861878,0.9516129032258065,0.9645776566757494,372.0
+LOC,0.9695652173913043,0.9330543933054394,0.9509594882729211,239.0
+WEBSITE,0.9668874172185431,0.9864864864864865,0.9765886287625418,148.0
+FAC,0.928,0.8992248062015504,0.9133858267716536,129.0
+LAW,1.0,1.0,1.0,91.0
+TIME,0.9156626506024096,0.8735632183908046,0.8941176470588236,87.0
+MONEY,0.9736842105263158,0.8809523809523809,0.925,42.0
+CURR,0.975609756097561,0.975609756097561,0.975609756097561,41.0
+LANGUAGE,0.875,0.8484848484848485,0.8615384615384616,33.0
+PERCENT,0.9642857142857143,0.8709677419354839,0.9152542372881356,31.0
+PRODUCT,1.0,0.9473684210526315,0.972972972972973,19.0
+QUANTITY,0.9090909090909091,1.0,0.9523809523809523,10.0
+UNIT,1.0,1.0,1.0,11.0
+accuracy,0.9793650793650793,0.9793650793650793,0.9793650793650793,0.9793650793650793
+macro avg,0.9672581856551346,0.9534679938651157,0.9599200012897606,12600.0
+weighted avg,0.9793254669511197,0.9793650793650793,0.9792974524068463,12600.0

metrics/test_overall_per_class.json

@@ -0,0 +1,141 @@
+{
+  "GPE": {
+    "precision": 0.9738970588235294,
+    "recall": 0.9843924191750278,
+    "f1-score": 0.9791166143041952,
+    "support": 2691.0
+  },
+  "ORG": {
+    "precision": 0.9807692307692307,
+    "recall": 0.9781872884543061,
+    "f1-score": 0.9794765580869893,
+    "support": 2659.0
+  },
+  "DATE": {
+    "precision": 0.9904378459989934,
+    "recall": 0.9954476479514416,
+    "f1-score": 0.992936427850656,
+    "support": 1977.0
+  },
+  "PERS": {
+    "precision": 0.9795348837209302,
+    "recall": 0.9831932773109243,
+    "f1-score": 0.9813606710158435,
+    "support": 1071.0
+  },
+  "NORP": {
+    "precision": 0.9765395894428153,
+    "recall": 0.9784524975514202,
+    "f1-score": 0.9774951076320939,
+    "support": 1021.0
+  },
+  "ORDINAL": {
+    "precision": 0.9820971867007673,
+    "recall": 0.9833546734955185,
+    "f1-score": 0.982725527831094,
+    "support": 781.0
+  },
+  "OCC": {
+    "precision": 0.9870466321243523,
+    "recall": 0.9896103896103896,
+    "f1-score": 0.9883268482490273,
+    "support": 770.0
+  },
+  "EVENT": {
+    "precision": 0.9864130434782609,
+    "recall": 0.9628647214854111,
+    "f1-score": 0.974496644295302,
+    "support": 377.0
+  },
+  "CARDINAL": {
+    "precision": 0.9779005524861878,
+    "recall": 0.9516129032258065,
+    "f1-score": 0.9645776566757494,
+    "support": 372.0
+  },
+  "LOC": {
+    "precision": 0.9695652173913043,
+    "recall": 0.9330543933054394,
+    "f1-score": 0.9509594882729211,
+    "support": 239.0
+  },
+  "WEBSITE": {
+    "precision": 0.9668874172185431,
+    "recall": 0.9864864864864865,
+    "f1-score": 0.9765886287625418,
+    "support": 148.0
+  },
+  "FAC": {
+    "precision": 0.928,
+    "recall": 0.8992248062015504,
+    "f1-score": 0.9133858267716536,
+    "support": 129.0
+  },
+  "LAW": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 91.0
+  },
+  "TIME": {
+    "precision": 0.9156626506024096,
+    "recall": 0.8735632183908046,
+    "f1-score": 0.8941176470588236,
+    "support": 87.0
+  },
+  "MONEY": {
+    "precision": 0.9736842105263158,
+    "recall": 0.8809523809523809,
+    "f1-score": 0.925,
+    "support": 42.0
+  },
+  "CURR": {
+    "precision": 0.975609756097561,
+    "recall": 0.975609756097561,
+    "f1-score": 0.975609756097561,
+    "support": 41.0
+  },
+  "LANGUAGE": {
+    "precision": 0.875,
+    "recall": 0.8484848484848485,
+    "f1-score": 0.8615384615384616,
+    "support": 33.0
+  },
+  "PERCENT": {
+    "precision": 0.9642857142857143,
+    "recall": 0.8709677419354839,
+    "f1-score": 0.9152542372881356,
+    "support": 31.0
+  },
+  "PRODUCT": {
+    "precision": 1.0,
+    "recall": 0.9473684210526315,
+    "f1-score": 0.972972972972973,
+    "support": 19.0
+  },
+  "QUANTITY": {
+    "precision": 0.9090909090909091,
+    "recall": 1.0,
+    "f1-score": 0.9523809523809523,
+    "support": 10.0
+  },
+  "UNIT": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 11.0
+  },
+  "accuracy": 0.9793650793650793,
+  "macro avg": {
+    "precision": 0.9672581856551346,
+    "recall": 0.9534679938651157,
+    "f1-score": 0.9599200012897606,
+    "support": 12600.0
+  },
+  "weighted avg": {
+    "precision": 0.9793254669511197,
+    "recall": 0.9793650793650793,
+    "f1-score": 0.9792974524068463,
+    "support": 12600.0
+  }
+}

metrics/test_seen_sentence_new_entity.json

@@ -0,0 +1,45 @@
+{
+  "subset": "seen_sentence_new_entity",
+  "rows": 11606,
+  "types_present": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "num_types_present": 21,
+  "accuracy": 0.9807858004480441,
+  "micro_precision": 0.9807858004480441,
+  "micro_recall": 0.9807858004480441,
+  "micro_f1": 0.9807858004480441,
+  "macro_precision_present_types": 0.970282742754196,
+  "macro_recall_present_types": 0.9589517514121707,
+  "macro_f1_present_types": 0.9641131075953279,
+  "macro_precision_all_21": 0.970282742754196,
+  "macro_recall_all_21": 0.9589517514121707,
+  "macro_f1_all_21": 0.9641131075953279,
+  "weighted_precision": 0.9807493542827415,
+  "weighted_recall": 0.9807858004480441,
+  "weighted_f1": 0.9807123412635281,
+  "split": "test",
+  "checkpoint": "best",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "generated_at_utc": "2026-07-12T18:56:34+00:00"
+}

metrics/test_seen_sentence_new_entity_per_class.csv

@@ -0,0 +1,25 @@
+,precision,recall,f1-score,support
+GPE,0.9750972762645914,0.987391646966115,0.9812059514487079,2538.0
+ORG,0.9816951850378034,0.9789682539682539,0.9803298231670972,2520.0
+DATE,0.9924852388620504,0.9956919763058697,0.9940860215053764,1857.0
+PERS,0.9796355841371919,0.9849137931034483,0.9822675980655562,928.0
+NORP,0.9788182831661093,0.9799107142857143,0.9793641940881205,896.0
+ORDINAL,0.9828080229226361,0.9814020028612304,0.9821045096635648,699.0
+OCC,0.9861878453038674,0.9916666666666667,0.9889196675900277,720.0
+EVENT,0.9859154929577465,0.9641873278236914,0.9749303621169917,363.0
+CARDINAL,0.9777070063694268,0.9504643962848297,0.9638932496075353,323.0
+LOC,0.9720930232558139,0.9330357142857143,0.9521640091116174,224.0
+WEBSITE,0.9791666666666666,0.9791666666666666,0.9791666666666666,96.0
+FAC,0.9310344827586207,0.8925619834710744,0.9113924050632911,121.0
+LAW,1.0,1.0,1.0,83.0
+TIME,0.9206349206349206,0.8529411764705882,0.8854961832061069,68.0
+MONEY,0.9736842105263158,0.9024390243902439,0.9367088607594937,41.0
+CURR,0.9743589743589743,1.0,0.987012987012987,38.0
+LANGUAGE,0.9230769230769231,0.96,0.9411764705882353,25.0
+PERCENT,0.9615384615384616,0.8620689655172413,0.9090909090909091,29.0
+PRODUCT,1.0,0.9411764705882353,0.9696969696969697,17.0
+QUANTITY,0.9,1.0,0.9473684210526315,9.0
+UNIT,1.0,1.0,1.0,11.0
+accuracy,0.9807858004480441,0.9807858004480441,0.9807858004480441,0.9807858004480441
+macro avg,0.970282742754196,0.9589517514121707,0.9641131075953279,11606.0
+weighted avg,0.9807493542827415,0.9807858004480441,0.9807123412635281,11606.0

metrics/test_seen_sentence_new_entity_per_class.json

@@ -0,0 +1,141 @@
+{
+  "GPE": {
+    "precision": 0.9750972762645914,
+    "recall": 0.987391646966115,
+    "f1-score": 0.9812059514487079,
+    "support": 2538.0
+  },
+  "ORG": {
+    "precision": 0.9816951850378034,
+    "recall": 0.9789682539682539,
+    "f1-score": 0.9803298231670972,
+    "support": 2520.0
+  },
+  "DATE": {
+    "precision": 0.9924852388620504,
+    "recall": 0.9956919763058697,
+    "f1-score": 0.9940860215053764,
+    "support": 1857.0
+  },
+  "PERS": {
+    "precision": 0.9796355841371919,
+    "recall": 0.9849137931034483,
+    "f1-score": 0.9822675980655562,
+    "support": 928.0
+  },
+  "NORP": {
+    "precision": 0.9788182831661093,
+    "recall": 0.9799107142857143,
+    "f1-score": 0.9793641940881205,
+    "support": 896.0
+  },
+  "ORDINAL": {
+    "precision": 0.9828080229226361,
+    "recall": 0.9814020028612304,
+    "f1-score": 0.9821045096635648,
+    "support": 699.0
+  },
+  "OCC": {
+    "precision": 0.9861878453038674,
+    "recall": 0.9916666666666667,
+    "f1-score": 0.9889196675900277,
+    "support": 720.0
+  },
+  "EVENT": {
+    "precision": 0.9859154929577465,
+    "recall": 0.9641873278236914,
+    "f1-score": 0.9749303621169917,
+    "support": 363.0
+  },
+  "CARDINAL": {
+    "precision": 0.9777070063694268,
+    "recall": 0.9504643962848297,
+    "f1-score": 0.9638932496075353,
+    "support": 323.0
+  },
+  "LOC": {
+    "precision": 0.9720930232558139,
+    "recall": 0.9330357142857143,
+    "f1-score": 0.9521640091116174,
+    "support": 224.0
+  },
+  "WEBSITE": {
+    "precision": 0.9791666666666666,
+    "recall": 0.9791666666666666,
+    "f1-score": 0.9791666666666666,
+    "support": 96.0
+  },
+  "FAC": {
+    "precision": 0.9310344827586207,
+    "recall": 0.8925619834710744,
+    "f1-score": 0.9113924050632911,
+    "support": 121.0
+  },
+  "LAW": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 83.0
+  },
+  "TIME": {
+    "precision": 0.9206349206349206,
+    "recall": 0.8529411764705882,
+    "f1-score": 0.8854961832061069,
+    "support": 68.0
+  },
+  "MONEY": {
+    "precision": 0.9736842105263158,
+    "recall": 0.9024390243902439,
+    "f1-score": 0.9367088607594937,
+    "support": 41.0
+  },
+  "CURR": {
+    "precision": 0.9743589743589743,
+    "recall": 1.0,
+    "f1-score": 0.987012987012987,
+    "support": 38.0
+  },
+  "LANGUAGE": {
+    "precision": 0.9230769230769231,
+    "recall": 0.96,
+    "f1-score": 0.9411764705882353,
+    "support": 25.0
+  },
+  "PERCENT": {
+    "precision": 0.9615384615384616,
+    "recall": 0.8620689655172413,
+    "f1-score": 0.9090909090909091,
+    "support": 29.0
+  },
+  "PRODUCT": {
+    "precision": 1.0,
+    "recall": 0.9411764705882353,
+    "f1-score": 0.9696969696969697,
+    "support": 17.0
+  },
+  "QUANTITY": {
+    "precision": 0.9,
+    "recall": 1.0,
+    "f1-score": 0.9473684210526315,
+    "support": 9.0
+  },
+  "UNIT": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 11.0
+  },
+  "accuracy": 0.9807858004480441,
+  "macro avg": {
+    "precision": 0.970282742754196,
+    "recall": 0.9589517514121707,
+    "f1-score": 0.9641131075953279,
+    "support": 11606.0
+  },
+  "weighted avg": {
+    "precision": 0.9807493542827415,
+    "recall": 0.9807858004480441,
+    "f1-score": 0.9807123412635281,
+    "support": 11606.0
+  }
+}

metrics/test_unseen_sentence.json

@@ -0,0 +1,44 @@
+{
+  "subset": "unseen_sentence",
+  "rows": 994,
+  "types_present": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY"
+  ],
+  "num_types_present": 20,
+  "accuracy": 0.9627766599597586,
+  "micro_precision": 0.9627766599597586,
+  "micro_recall": 0.9627766599597586,
+  "micro_f1": 0.9627766599597586,
+  "macro_precision_present_types": 0.9052965403310622,
+  "macro_recall_present_types": 0.886274373082116,
+  "macro_f1_present_types": 0.8931735856196752,
+  "macro_precision_all_21": 0.8621871812676783,
+  "macro_recall_all_21": 0.8440708315067772,
+  "macro_f1_all_21": 0.8506415101139764,
+  "weighted_precision": 0.9615439764829642,
+  "weighted_recall": 0.9627766599597586,
+  "weighted_f1": 0.9617803672434098,
+  "split": "test",
+  "checkpoint": "best",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "generated_at_utc": "2026-07-12T18:56:34+00:00"
+}

metrics/test_unseen_sentence_per_class.csv

@@ -0,0 +1,25 @@
+,precision,recall,f1-score,support
+GPE,0.9533333333333334,0.934640522875817,0.9438943894389439,153.0
+ORG,0.9640287769784173,0.9640287769784173,0.9640287769784173,139.0
+DATE,0.9596774193548387,0.9916666666666667,0.9754098360655737,120.0
+PERS,0.9788732394366197,0.972027972027972,0.9754385964912281,143.0
+NORP,0.9603174603174603,0.968,0.9641434262948207,125.0
+ORDINAL,0.9761904761904762,1.0,0.9879518072289156,82.0
+OCC,1.0,0.96,0.9795918367346939,50.0
+EVENT,1.0,0.9285714285714286,0.9629629629629629,14.0
+CARDINAL,0.9791666666666666,0.9591836734693877,0.9690721649484536,49.0
+LOC,0.9333333333333333,0.9333333333333333,0.9333333333333333,15.0
+WEBSITE,0.9454545454545454,1.0,0.9719626168224299,52.0
+FAC,0.8888888888888888,1.0,0.9411764705882353,8.0
+LAW,1.0,1.0,1.0,8.0
+TIME,0.9,0.9473684210526315,0.9230769230769231,19.0
+MONEY,0.0,0.0,0.0,1.0
+CURR,1.0,0.6666666666666666,0.8,3.0
+LANGUAGE,0.6666666666666666,0.5,0.5714285714285714,8.0
+PERCENT,1.0,1.0,1.0,2.0
+PRODUCT,1.0,1.0,1.0,2.0
+QUANTITY,1.0,1.0,1.0,1.0
+UNIT,0.0,0.0,0.0,0.0
+accuracy,0.9627766599597586,0.9627766599597586,0.9627766599597586,0.9627766599597586
+macro avg,0.8621871812676783,0.8440708315067772,0.8506415101139764,994.0
+weighted avg,0.9615439764829642,0.9627766599597586,0.9617803672434098,994.0

metrics/test_unseen_sentence_per_class.json

@@ -0,0 +1,141 @@
+{
+  "GPE": {
+    "precision": 0.9533333333333334,
+    "recall": 0.934640522875817,
+    "f1-score": 0.9438943894389439,
+    "support": 153.0
+  },
+  "ORG": {
+    "precision": 0.9640287769784173,
+    "recall": 0.9640287769784173,
+    "f1-score": 0.9640287769784173,
+    "support": 139.0
+  },
+  "DATE": {
+    "precision": 0.9596774193548387,
+    "recall": 0.9916666666666667,
+    "f1-score": 0.9754098360655737,
+    "support": 120.0
+  },
+  "PERS": {
+    "precision": 0.9788732394366197,
+    "recall": 0.972027972027972,
+    "f1-score": 0.9754385964912281,
+    "support": 143.0
+  },
+  "NORP": {
+    "precision": 0.9603174603174603,
+    "recall": 0.968,
+    "f1-score": 0.9641434262948207,
+    "support": 125.0
+  },
+  "ORDINAL": {
+    "precision": 0.9761904761904762,
+    "recall": 1.0,
+    "f1-score": 0.9879518072289156,
+    "support": 82.0
+  },
+  "OCC": {
+    "precision": 1.0,
+    "recall": 0.96,
+    "f1-score": 0.9795918367346939,
+    "support": 50.0
+  },
+  "EVENT": {
+    "precision": 1.0,
+    "recall": 0.9285714285714286,
+    "f1-score": 0.9629629629629629,
+    "support": 14.0
+  },
+  "CARDINAL": {
+    "precision": 0.9791666666666666,
+    "recall": 0.9591836734693877,
+    "f1-score": 0.9690721649484536,
+    "support": 49.0
+  },
+  "LOC": {
+    "precision": 0.9333333333333333,
+    "recall": 0.9333333333333333,
+    "f1-score": 0.9333333333333333,
+    "support": 15.0
+  },
+  "WEBSITE": {
+    "precision": 0.9454545454545454,
+    "recall": 1.0,
+    "f1-score": 0.9719626168224299,
+    "support": 52.0
+  },
+  "FAC": {
+    "precision": 0.8888888888888888,
+    "recall": 1.0,
+    "f1-score": 0.9411764705882353,
+    "support": 8.0
+  },
+  "LAW": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 8.0
+  },
+  "TIME": {
+    "precision": 0.9,
+    "recall": 0.9473684210526315,
+    "f1-score": 0.9230769230769231,
+    "support": 19.0
+  },
+  "MONEY": {
+    "precision": 0.0,
+    "recall": 0.0,
+    "f1-score": 0.0,
+    "support": 1.0
+  },
+  "CURR": {
+    "precision": 1.0,
+    "recall": 0.6666666666666666,
+    "f1-score": 0.8,
+    "support": 3.0
+  },
+  "LANGUAGE": {
+    "precision": 0.6666666666666666,
+    "recall": 0.5,
+    "f1-score": 0.5714285714285714,
+    "support": 8.0
+  },
+  "PERCENT": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 2.0
+  },
+  "PRODUCT": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 2.0
+  },
+  "QUANTITY": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 1.0
+  },
+  "UNIT": {
+    "precision": 0.0,
+    "recall": 0.0,
+    "f1-score": 0.0,
+    "support": 0.0
+  },
+  "accuracy": 0.9627766599597586,
+  "macro avg": {
+    "precision": 0.8621871812676783,
+    "recall": 0.8440708315067772,
+    "f1-score": 0.8506415101139764,
+    "support": 994.0
+  },
+  "weighted avg": {
+    "precision": 0.9615439764829642,
+    "recall": 0.9627766599597586,
+    "f1-score": 0.9617803672434098,
+    "support": 994.0
+  }
+}

metrics/validation_overall.json

@@ -0,0 +1,45 @@
+{
+  "subset": "overall",
+  "rows": 12600,
+  "types_present": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "num_types_present": 21,
+  "accuracy": 0.9778571428571429,
+  "micro_precision": 0.9778571428571429,
+  "micro_recall": 0.9778571428571429,
+  "micro_f1": 0.9778571428571429,
+  "macro_precision_present_types": 0.9725024753902682,
+  "macro_recall_present_types": 0.9475659044801065,
+  "macro_f1_present_types": 0.9587281193044483,
+  "macro_precision_all_21": 0.9725024753902682,
+  "macro_recall_all_21": 0.9475659044801065,
+  "macro_f1_all_21": 0.9587281193044483,
+  "weighted_precision": 0.977879248064854,
+  "weighted_recall": 0.9778571428571429,
+  "weighted_f1": 0.9777906938936711,
+  "split": "validation",
+  "checkpoint": "best",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "generated_at_utc": "2026-07-12T18:55:47+00:00"
+}

metrics/validation_overall_per_class.csv

@@ -0,0 +1,25 @@
+,precision,recall,f1-score,support
+GPE,0.977432482426933,0.9821561338289962,0.9797886148711292,2690.0
+ORG,0.9728321548195012,0.9827067669172932,0.9777445296427903,2660.0
+DATE,0.9934310257705912,0.9949392712550608,0.9941845764854614,1976.0
+PERS,0.9738317757009346,0.9720149253731343,0.972922502334267,1072.0
+NORP,0.9750499001996008,0.9559686888454012,0.9654150197628458,1022.0
+ORDINAL,0.9795134443021767,0.9807692307692307,0.9801409352978859,780.0
+OCC,0.981888745148771,0.9869960988296489,0.9844357976653697,769.0
+EVENT,0.9673024523160763,0.9416445623342176,0.9543010752688172,377.0
+CARDINAL,0.9733333333333334,0.9785522788203753,0.9759358288770054,373.0
+LOC,0.9745762711864406,0.9623430962343096,0.968421052631579,239.0
+WEBSITE,0.9865771812080537,0.9932432432432432,0.98989898989899,148.0
+FAC,0.9166666666666666,0.9307692307692308,0.9236641221374046,130.0
+LAW,1.0,1.0,1.0,90.0
+TIME,0.9743589743589743,0.8735632183908046,0.9212121212121213,87.0
+MONEY,0.9333333333333333,1.0,0.9655172413793104,42.0
+CURR,1.0,0.9512195121951219,0.975,41.0
+LANGUAGE,0.9333333333333333,0.8484848484848485,0.8888888888888888,33.0
+PERCENT,1.0,0.9032258064516129,0.9491525423728814,31.0
+PRODUCT,1.0,0.8421052631578947,0.9142857142857143,19.0
+QUANTITY,1.0,0.8181818181818182,0.9,11.0
+UNIT,0.9090909090909091,1.0,0.9523809523809523,10.0
+accuracy,0.9778571428571429,0.9778571428571429,0.9778571428571429,0.9778571428571429
+macro avg,0.9725024753902682,0.9475659044801065,0.9587281193044483,12600.0
+weighted avg,0.977879248064854,0.9778571428571429,0.9777906938936711,12600.0

metrics/validation_overall_per_class.json

@@ -0,0 +1,141 @@
+{
+  "GPE": {
+    "precision": 0.977432482426933,
+    "recall": 0.9821561338289962,
+    "f1-score": 0.9797886148711292,
+    "support": 2690.0
+  },
+  "ORG": {
+    "precision": 0.9728321548195012,
+    "recall": 0.9827067669172932,
+    "f1-score": 0.9777445296427903,
+    "support": 2660.0
+  },
+  "DATE": {
+    "precision": 0.9934310257705912,
+    "recall": 0.9949392712550608,
+    "f1-score": 0.9941845764854614,
+    "support": 1976.0
+  },
+  "PERS": {
+    "precision": 0.9738317757009346,
+    "recall": 0.9720149253731343,
+    "f1-score": 0.972922502334267,
+    "support": 1072.0
+  },
+  "NORP": {
+    "precision": 0.9750499001996008,
+    "recall": 0.9559686888454012,
+    "f1-score": 0.9654150197628458,
+    "support": 1022.0
+  },
+  "ORDINAL": {
+    "precision": 0.9795134443021767,
+    "recall": 0.9807692307692307,
+    "f1-score": 0.9801409352978859,
+    "support": 780.0
+  },
+  "OCC": {
+    "precision": 0.981888745148771,
+    "recall": 0.9869960988296489,
+    "f1-score": 0.9844357976653697,
+    "support": 769.0
+  },
+  "EVENT": {
+    "precision": 0.9673024523160763,
+    "recall": 0.9416445623342176,
+    "f1-score": 0.9543010752688172,
+    "support": 377.0
+  },
+  "CARDINAL": {
+    "precision": 0.9733333333333334,
+    "recall": 0.9785522788203753,
+    "f1-score": 0.9759358288770054,
+    "support": 373.0
+  },
+  "LOC": {
+    "precision": 0.9745762711864406,
+    "recall": 0.9623430962343096,
+    "f1-score": 0.968421052631579,
+    "support": 239.0
+  },
+  "WEBSITE": {
+    "precision": 0.9865771812080537,
+    "recall": 0.9932432432432432,
+    "f1-score": 0.98989898989899,
+    "support": 148.0
+  },
+  "FAC": {
+    "precision": 0.9166666666666666,
+    "recall": 0.9307692307692308,
+    "f1-score": 0.9236641221374046,
+    "support": 130.0
+  },
+  "LAW": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 90.0
+  },
+  "TIME": {
+    "precision": 0.9743589743589743,
+    "recall": 0.8735632183908046,
+    "f1-score": 0.9212121212121213,
+    "support": 87.0
+  },
+  "MONEY": {
+    "precision": 0.9333333333333333,
+    "recall": 1.0,
+    "f1-score": 0.9655172413793104,
+    "support": 42.0
+  },
+  "CURR": {
+    "precision": 1.0,
+    "recall": 0.9512195121951219,
+    "f1-score": 0.975,
+    "support": 41.0
+  },
+  "LANGUAGE": {
+    "precision": 0.9333333333333333,
+    "recall": 0.8484848484848485,
+    "f1-score": 0.8888888888888888,
+    "support": 33.0
+  },
+  "PERCENT": {
+    "precision": 1.0,
+    "recall": 0.9032258064516129,
+    "f1-score": 0.9491525423728814,
+    "support": 31.0
+  },
+  "PRODUCT": {
+    "precision": 1.0,
+    "recall": 0.8421052631578947,
+    "f1-score": 0.9142857142857143,
+    "support": 19.0
+  },
+  "QUANTITY": {
+    "precision": 1.0,
+    "recall": 0.8181818181818182,
+    "f1-score": 0.9,
+    "support": 11.0
+  },
+  "UNIT": {
+    "precision": 0.9090909090909091,
+    "recall": 1.0,
+    "f1-score": 0.9523809523809523,
+    "support": 10.0
+  },
+  "accuracy": 0.9778571428571429,
+  "macro avg": {
+    "precision": 0.9725024753902682,
+    "recall": 0.9475659044801065,
+    "f1-score": 0.9587281193044483,
+    "support": 12600.0
+  },
+  "weighted avg": {
+    "precision": 0.977879248064854,
+    "recall": 0.9778571428571429,
+    "f1-score": 0.9777906938936711,
+    "support": 12600.0
+  }
+}

metrics/validation_seen_sentence_new_entity.json

@@ -0,0 +1,45 @@
+{
+  "subset": "seen_sentence_new_entity",
+  "rows": 11661,
+  "types_present": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "QUANTITY",
+    "UNIT"
+  ],
+  "num_types_present": 21,
+  "accuracy": 0.9804476459994854,
+  "micro_precision": 0.9804476459994854,
+  "micro_recall": 0.9804476459994854,
+  "micro_f1": 0.9804476459994854,
+  "macro_precision_present_types": 0.9751134735007793,
+  "macro_recall_present_types": 0.9537701979295099,
+  "macro_f1_present_types": 0.963262123278673,
+  "macro_precision_all_21": 0.9751134735007793,
+  "macro_recall_all_21": 0.9537701979295099,
+  "macro_f1_all_21": 0.963262123278673,
+  "weighted_precision": 0.980497781530163,
+  "weighted_recall": 0.9804476459994854,
+  "weighted_f1": 0.9804033222521132,
+  "split": "validation",
+  "checkpoint": "best",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "generated_at_utc": "2026-07-12T18:55:47+00:00"
+}

metrics/validation_seen_sentence_new_entity_per_class.csv

@@ -0,0 +1,25 @@
+,precision,recall,f1-score,support
+GPE,0.9795677799607073,0.9849861714737258,0.9822695035460993,2531.0
+ORG,0.9750778816199377,0.983503534956795,0.9792725850606179,2546.0
+DATE,0.9946062567421791,0.9956803455723542,0.9951430113329736,1852.0
+PERS,0.9804347826086957,0.9730312837108953,0.9767190037899296,927.0
+NORP,0.9789356984478935,0.9587404994571118,0.9687328579264948,921.0
+ORDINAL,0.9854862119013063,0.9826338639652678,0.9840579710144928,691.0
+OCC,0.98224043715847,0.9930939226519337,0.9876373626373627,724.0
+EVENT,0.9716713881019831,0.9475138121546961,0.9594405594405594,362.0
+CARDINAL,0.9724770642201835,0.9814814814814815,0.9769585253456221,324.0
+LOC,0.9772727272727273,0.9684684684684685,0.9728506787330317,222.0
+WEBSITE,0.990990990990991,0.990990990990991,0.990990990990991,111.0
+FAC,0.9083333333333333,0.9316239316239316,0.919831223628692,117.0
+LAW,1.0,1.0,1.0,83.0
+TIME,0.9855072463768116,0.9066666666666666,0.9444444444444444,75.0
+MONEY,0.9318181818181818,1.0,0.9647058823529412,41.0
+CURR,1.0,0.9743589743589743,0.987012987012987,39.0
+LANGUAGE,0.9629629629629629,0.8666666666666667,0.9122807017543859,30.0
+PERCENT,1.0,0.896551724137931,0.9454545454545454,29.0
+PRODUCT,1.0,0.875,0.9333333333333333,16.0
+QUANTITY,1.0,0.8181818181818182,0.9,11.0
+UNIT,0.9,1.0,0.9473684210526315,9.0
+accuracy,0.9804476459994854,0.9804476459994854,0.9804476459994854,0.9804476459994854
+macro avg,0.9751134735007793,0.9537701979295099,0.963262123278673,11661.0
+weighted avg,0.980497781530163,0.9804476459994854,0.9804033222521132,11661.0

metrics/validation_seen_sentence_new_entity_per_class.json

@@ -0,0 +1,141 @@
+{
+  "GPE": {
+    "precision": 0.9795677799607073,
+    "recall": 0.9849861714737258,
+    "f1-score": 0.9822695035460993,
+    "support": 2531.0
+  },
+  "ORG": {
+    "precision": 0.9750778816199377,
+    "recall": 0.983503534956795,
+    "f1-score": 0.9792725850606179,
+    "support": 2546.0
+  },
+  "DATE": {
+    "precision": 0.9946062567421791,
+    "recall": 0.9956803455723542,
+    "f1-score": 0.9951430113329736,
+    "support": 1852.0
+  },
+  "PERS": {
+    "precision": 0.9804347826086957,
+    "recall": 0.9730312837108953,
+    "f1-score": 0.9767190037899296,
+    "support": 927.0
+  },
+  "NORP": {
+    "precision": 0.9789356984478935,
+    "recall": 0.9587404994571118,
+    "f1-score": 0.9687328579264948,
+    "support": 921.0
+  },
+  "ORDINAL": {
+    "precision": 0.9854862119013063,
+    "recall": 0.9826338639652678,
+    "f1-score": 0.9840579710144928,
+    "support": 691.0
+  },
+  "OCC": {
+    "precision": 0.98224043715847,
+    "recall": 0.9930939226519337,
+    "f1-score": 0.9876373626373627,
+    "support": 724.0
+  },
+  "EVENT": {
+    "precision": 0.9716713881019831,
+    "recall": 0.9475138121546961,
+    "f1-score": 0.9594405594405594,
+    "support": 362.0
+  },
+  "CARDINAL": {
+    "precision": 0.9724770642201835,
+    "recall": 0.9814814814814815,
+    "f1-score": 0.9769585253456221,
+    "support": 324.0
+  },
+  "LOC": {
+    "precision": 0.9772727272727273,
+    "recall": 0.9684684684684685,
+    "f1-score": 0.9728506787330317,
+    "support": 222.0
+  },
+  "WEBSITE": {
+    "precision": 0.990990990990991,
+    "recall": 0.990990990990991,
+    "f1-score": 0.990990990990991,
+    "support": 111.0
+  },
+  "FAC": {
+    "precision": 0.9083333333333333,
+    "recall": 0.9316239316239316,
+    "f1-score": 0.919831223628692,
+    "support": 117.0
+  },
+  "LAW": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 83.0
+  },
+  "TIME": {
+    "precision": 0.9855072463768116,
+    "recall": 0.9066666666666666,
+    "f1-score": 0.9444444444444444,
+    "support": 75.0
+  },
+  "MONEY": {
+    "precision": 0.9318181818181818,
+    "recall": 1.0,
+    "f1-score": 0.9647058823529412,
+    "support": 41.0
+  },
+  "CURR": {
+    "precision": 1.0,
+    "recall": 0.9743589743589743,
+    "f1-score": 0.987012987012987,
+    "support": 39.0
+  },
+  "LANGUAGE": {
+    "precision": 0.9629629629629629,
+    "recall": 0.8666666666666667,
+    "f1-score": 0.9122807017543859,
+    "support": 30.0
+  },
+  "PERCENT": {
+    "precision": 1.0,
+    "recall": 0.896551724137931,
+    "f1-score": 0.9454545454545454,
+    "support": 29.0
+  },
+  "PRODUCT": {
+    "precision": 1.0,
+    "recall": 0.875,
+    "f1-score": 0.9333333333333333,
+    "support": 16.0
+  },
+  "QUANTITY": {
+    "precision": 1.0,
+    "recall": 0.8181818181818182,
+    "f1-score": 0.9,
+    "support": 11.0
+  },
+  "UNIT": {
+    "precision": 0.9,
+    "recall": 1.0,
+    "f1-score": 0.9473684210526315,
+    "support": 9.0
+  },
+  "accuracy": 0.9804476459994854,
+  "macro avg": {
+    "precision": 0.9751134735007793,
+    "recall": 0.9537701979295099,
+    "f1-score": 0.963262123278673,
+    "support": 11661.0
+  },
+  "weighted avg": {
+    "precision": 0.980497781530163,
+    "recall": 0.9804476459994854,
+    "f1-score": 0.9804033222521132,
+    "support": 11661.0
+  }
+}

metrics/validation_unseen_sentence.json

@@ -0,0 +1,44 @@
+{
+  "subset": "unseen_sentence",
+  "rows": 939,
+  "types_present": [
+    "GPE",
+    "ORG",
+    "DATE",
+    "PERS",
+    "NORP",
+    "ORDINAL",
+    "OCC",
+    "EVENT",
+    "CARDINAL",
+    "LOC",
+    "WEBSITE",
+    "FAC",
+    "LAW",
+    "TIME",
+    "MONEY",
+    "CURR",
+    "LANGUAGE",
+    "PERCENT",
+    "PRODUCT",
+    "UNIT"
+  ],
+  "num_types_present": 20,
+  "accuracy": 0.9456869009584664,
+  "micro_precision": 0.9456869009584664,
+  "micro_recall": 0.9456869009584664,
+  "micro_f1": 0.9456869009584664,
+  "macro_precision_present_types": 0.9465091355300324,
+  "macro_recall_present_types": 0.8850947519406029,
+  "macro_f1_present_types": 0.9088413774156396,
+  "macro_precision_all_21": 0.9014372719333641,
+  "macro_recall_all_21": 0.8429473828005742,
+  "macro_f1_all_21": 0.8655632165863235,
+  "weighted_precision": 0.9458822212425712,
+  "weighted_recall": 0.9456869009584664,
+  "weighted_f1": 0.945010776949412,
+  "split": "validation",
+  "checkpoint": "best",
+  "model_repo_id": "U4RASD/TypePredictor",
+  "generated_at_utc": "2026-07-12T18:55:47+00:00"
+}

metrics/validation_unseen_sentence_per_class.csv

@@ -0,0 +1,25 @@
+,precision,recall,f1-score,support
+GPE,0.9430379746835443,0.9371069182389937,0.9400630914826499,159.0
+ORG,0.9243697478991597,0.9649122807017544,0.944206008583691,114.0
+DATE,0.976,0.9838709677419355,0.9799196787148594,124.0
+PERS,0.9333333333333333,0.9655172413793104,0.9491525423728814,145.0
+NORP,0.94,0.9306930693069307,0.9353233830845771,101.0
+ORDINAL,0.9347826086956522,0.9662921348314607,0.9502762430939227,89.0
+OCC,0.975609756097561,0.8888888888888888,0.9302325581395349,45.0
+EVENT,0.8571428571428571,0.8,0.8275862068965517,15.0
+CARDINAL,0.9791666666666666,0.9591836734693877,0.9690721649484536,49.0
+LOC,0.9375,0.8823529411764706,0.9090909090909091,17.0
+WEBSITE,0.9736842105263158,1.0,0.9866666666666667,37.0
+FAC,1.0,0.9230769230769231,0.96,13.0
+LAW,1.0,1.0,1.0,7.0
+TIME,0.8888888888888888,0.6666666666666666,0.7619047619047619,12.0
+MONEY,1.0,1.0,1.0,1.0
+CURR,1.0,0.5,0.6666666666666666,2.0
+LANGUAGE,0.6666666666666666,0.6666666666666666,0.6666666666666666,3.0
+PERCENT,1.0,1.0,1.0,2.0
+PRODUCT,1.0,0.6666666666666666,0.8,3.0
+QUANTITY,0.0,0.0,0.0,0.0
+UNIT,1.0,1.0,1.0,1.0
+accuracy,0.9456869009584664,0.9456869009584664,0.9456869009584664,0.9456869009584664
+macro avg,0.9014372719333641,0.8429473828005742,0.8655632165863235,939.0
+weighted avg,0.9458822212425712,0.9456869009584664,0.945010776949412,939.0

metrics/validation_unseen_sentence_per_class.json

@@ -0,0 +1,141 @@
+{
+  "GPE": {
+    "precision": 0.9430379746835443,
+    "recall": 0.9371069182389937,
+    "f1-score": 0.9400630914826499,
+    "support": 159.0
+  },
+  "ORG": {
+    "precision": 0.9243697478991597,
+    "recall": 0.9649122807017544,
+    "f1-score": 0.944206008583691,
+    "support": 114.0
+  },
+  "DATE": {
+    "precision": 0.976,
+    "recall": 0.9838709677419355,
+    "f1-score": 0.9799196787148594,
+    "support": 124.0
+  },
+  "PERS": {
+    "precision": 0.9333333333333333,
+    "recall": 0.9655172413793104,
+    "f1-score": 0.9491525423728814,
+    "support": 145.0
+  },
+  "NORP": {
+    "precision": 0.94,
+    "recall": 0.9306930693069307,
+    "f1-score": 0.9353233830845771,
+    "support": 101.0
+  },
+  "ORDINAL": {
+    "precision": 0.9347826086956522,
+    "recall": 0.9662921348314607,
+    "f1-score": 0.9502762430939227,
+    "support": 89.0
+  },
+  "OCC": {
+    "precision": 0.975609756097561,
+    "recall": 0.8888888888888888,
+    "f1-score": 0.9302325581395349,
+    "support": 45.0
+  },
+  "EVENT": {
+    "precision": 0.8571428571428571,
+    "recall": 0.8,
+    "f1-score": 0.8275862068965517,
+    "support": 15.0
+  },
+  "CARDINAL": {
+    "precision": 0.9791666666666666,
+    "recall": 0.9591836734693877,
+    "f1-score": 0.9690721649484536,
+    "support": 49.0
+  },
+  "LOC": {
+    "precision": 0.9375,
+    "recall": 0.8823529411764706,
+    "f1-score": 0.9090909090909091,
+    "support": 17.0
+  },
+  "WEBSITE": {
+    "precision": 0.9736842105263158,
+    "recall": 1.0,
+    "f1-score": 0.9866666666666667,
+    "support": 37.0
+  },
+  "FAC": {
+    "precision": 1.0,
+    "recall": 0.9230769230769231,
+    "f1-score": 0.96,
+    "support": 13.0
+  },
+  "LAW": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 7.0
+  },
+  "TIME": {
+    "precision": 0.8888888888888888,
+    "recall": 0.6666666666666666,
+    "f1-score": 0.7619047619047619,
+    "support": 12.0
+  },
+  "MONEY": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 1.0
+  },
+  "CURR": {
+    "precision": 1.0,
+    "recall": 0.5,
+    "f1-score": 0.6666666666666666,
+    "support": 2.0
+  },
+  "LANGUAGE": {
+    "precision": 0.6666666666666666,
+    "recall": 0.6666666666666666,
+    "f1-score": 0.6666666666666666,
+    "support": 3.0
+  },
+  "PERCENT": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 2.0
+  },
+  "PRODUCT": {
+    "precision": 1.0,
+    "recall": 0.6666666666666666,
+    "f1-score": 0.8,
+    "support": 3.0
+  },
+  "QUANTITY": {
+    "precision": 0.0,
+    "recall": 0.0,
+    "f1-score": 0.0,
+    "support": 0.0
+  },
+  "UNIT": {
+    "precision": 1.0,
+    "recall": 1.0,
+    "f1-score": 1.0,
+    "support": 1.0
+  },
+  "accuracy": 0.9456869009584664,
+  "macro avg": {
+    "precision": 0.9014372719333641,
+    "recall": 0.8429473828005742,
+    "f1-score": 0.8655632165863235,
+    "support": 939.0
+  },
+  "weighted avg": {
+    "precision": 0.9458822212425712,
+    "recall": 0.9456869009584664,
+    "f1-score": 0.945010776949412,
+    "support": 939.0
+  }
+}

modeling_type_predictor.py

@@ -0,0 +1,104 @@
+# -*- coding: utf-8 -*-
+"""Standalone loader for U4RASD/TypePredictor."""
+
+import json
+import types
+from pathlib import Path
+from typing import Optional, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from huggingface_hub import hf_hub_download
+from transformers import AutoModel, AutoTokenizer, PreTrainedTokenizerFast
+
+
+def _patch_neobert_embeddings_for_resize(encoder: nn.Module) -> nn.Module:
+    old_vocab_size = int(encoder.config.vocab_size)
+    parent = None
+    name = None
+    layer = None
+    for module in encoder.modules():
+        for child_name, child in module.named_children():
+            if isinstance(child, nn.Embedding) and child.num_embeddings == old_vocab_size:
+                parent = module
+                name = child_name
+                layer = child
+                break
+        if layer is not None:
+            break
+    if parent is None or name is None:
+        raise RuntimeError(f"Could not find token embedding layer with vocab size {old_vocab_size}.")
+
+    def get_input_embeddings(self):
+        return getattr(parent, name)
+
+    def set_input_embeddings(self, new_embeddings):
+        setattr(parent, name, new_embeddings)
+
+    encoder.get_input_embeddings = types.MethodType(get_input_embeddings, encoder)
+    encoder.set_input_embeddings = types.MethodType(set_input_embeddings, encoder)
+    return encoder
+
+
+class NeoAraBERTTypePredictor(nn.Module):
+    def __init__(self, base_model: str, tokenizer_size: int, num_labels: int = 21, dropout: float = 0.1):
+        super().__init__()
+        self.encoder = AutoModel.from_pretrained(base_model, trust_remote_code=True)
+        self.encoder = _patch_neobert_embeddings_for_resize(self.encoder)
+        try:
+            self.encoder.resize_token_embeddings(tokenizer_size, mean_resizing=False)
+        except TypeError:
+            self.encoder.resize_token_embeddings(tokenizer_size)
+        self.encoder.config.vocab_size = int(tokenizer_size)
+        self.hidden_size = int(self.encoder.config.hidden_size)
+        self.dropout = nn.Dropout(float(dropout))
+        self.classifier = nn.Linear(self.hidden_size, int(num_labels))
+
+    def forward(self, input_ids=None, attention_mask=None, labels=None):
+        outputs = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
+        hidden = outputs.last_hidden_state if hasattr(outputs, "last_hidden_state") else outputs[0]
+        logits = self.classifier(self.dropout(hidden[:, 0, :]))
+        result = {"logits": logits}
+        if labels is not None:
+            result["loss"] = F.cross_entropy(logits, labels.long())
+        return result
+
+    @classmethod
+    def from_pretrained(
+        cls,
+        model_id_or_path: Union[str, Path],
+        token: Optional[str] = None,
+        map_location: str = "cpu",
+    ):
+        model_id_or_path = str(model_id_or_path)
+        local_path = Path(model_id_or_path)
+        if local_path.exists():
+            config_path = local_path / "type_predictor_config.json"
+            weights_path = local_path / "pytorch_model.bin"
+            tokenizer_source = model_id_or_path
+        else:
+            config_path = Path(hf_hub_download(model_id_or_path, "type_predictor_config.json", token=token))
+            weights_path = Path(hf_hub_download(model_id_or_path, "pytorch_model.bin", token=token))
+            tokenizer_source = model_id_or_path
+
+        config = json.loads(config_path.read_text(encoding="utf-8"))
+        try:
+            tokenizer = AutoTokenizer.from_pretrained(
+                tokenizer_source, use_fast=True, trust_remote_code=False, token=token
+            )
+        except Exception:
+            tokenizer = PreTrainedTokenizerFast.from_pretrained(
+                tokenizer_source, trust_remote_code=False, token=token
+            )
+        if tokenizer is True or tokenizer is False:
+            raise RuntimeError("Tokenizer unexpectedly loaded as a boolean.")
+        model = cls(
+            base_model=config["base_model"],
+            tokenizer_size=len(tokenizer),
+            num_labels=config["num_labels"],
+            dropout=config["dropout"],
+        )
+        state_dict = torch.load(weights_path, map_location=map_location, weights_only=False)
+        model.load_state_dict(state_dict, strict=True)
+        return model, tokenizer, config

predictions/test_predictions.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcf7973d1e70244cde5b8639296e0150392d176f3150037dadfe01ddeaa88a54
+size 11819131