Text Classification
Transformers
Safetensors
English
deberta-v2
scientific-claim-verification
scifact
evidence-aware-verification
retrieval-aware-training
text-embeddings-inference
Instructions to use rishhh/verisci-claim-verifier-dense-adapted-seed123 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use rishhh/verisci-claim-verifier-dense-adapted-seed123 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="rishhh/verisci-claim-verifier-dense-adapted-seed123")# Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("rishhh/verisci-claim-verifier-dense-adapted-seed123") model = AutoModelForSequenceClassification.from_pretrained("rishhh/verisci-claim-verifier-dense-adapted-seed123", device_map="auto") - Notebooks
- Google Colab
- Kaggle
| { | |
| "datasets": { | |
| "hard_nei_source": "andreiaalexa/scifact-relevance-pairs title/train and BM25-retrieved non-cited SciFact train documents", | |
| "test_policy": "Validation rows with train-overlapping normalized claims are removed; remaining validation rows are split by normalized claim key into validation and derived test.", | |
| "verifier_primary": "allenai/scifact_entailment train plus allenai/scifact train retrieval-aware snippets" | |
| }, | |
| "full_scifact_validation": { | |
| "device": "cuda", | |
| "end_to_end": { | |
| "top1": { | |
| "accuracy": 0.5711111111111111, | |
| "confusion_matrix": { | |
| "NOT_ENOUGH_INFO": { | |
| "NOT_ENOUGH_INFO": 82, | |
| "REFUTES": 7, | |
| "SUPPORTS": 23 | |
| }, | |
| "REFUTES": { | |
| "NOT_ENOUGH_INFO": 67, | |
| "REFUTES": 47, | |
| "SUPPORTS": 8 | |
| }, | |
| "SUPPORTS": { | |
| "NOT_ENOUGH_INFO": 68, | |
| "REFUTES": 20, | |
| "SUPPORTS": 128 | |
| } | |
| }, | |
| "examples": 450, | |
| "gold_label_distribution": { | |
| "NOT_ENOUGH_INFO": 112, | |
| "REFUTES": 122, | |
| "SUPPORTS": 216 | |
| }, | |
| "macro_f1": 0.5535795821874849, | |
| "per_class": { | |
| "NOT_ENOUGH_INFO": { | |
| "f1-score": 0.49848024316109424, | |
| "precision": 0.3778801843317972, | |
| "recall": 0.7321428571428571, | |
| "support": 112.0 | |
| }, | |
| "REFUTES": { | |
| "f1-score": 0.47959183673469385, | |
| "precision": 0.6351351351351351, | |
| "recall": 0.38524590163934425, | |
| "support": 122.0 | |
| }, | |
| "SUPPORTS": { | |
| "f1-score": 0.6826666666666666, | |
| "precision": 0.8050314465408805, | |
| "recall": 0.5925925925925926, | |
| "support": 216.0 | |
| }, | |
| "accuracy": 0.5711111111111111, | |
| "macro avg": { | |
| "f1-score": 0.5535795821874849, | |
| "precision": 0.606015588669271, | |
| "recall": 0.5699937837915979, | |
| "support": 450.0 | |
| }, | |
| "weighted avg": { | |
| "f1-score": 0.5817688695903894, | |
| "precision": 0.6526574657432843, | |
| "recall": 0.5711111111111111, | |
| "support": 450.0 | |
| } | |
| }, | |
| "predicted_label_distribution": { | |
| "NOT_ENOUGH_INFO": 217, | |
| "REFUTES": 74, | |
| "SUPPORTS": 159 | |
| }, | |
| "strategy": "top1", | |
| "supported_refuted_accuracy_when_evidence_retrieved_at_5": 0.6415094339622641, | |
| "supported_refuted_examples_with_evidence_retrieved_at_5": 265 | |
| }, | |
| "top5_decisive": { | |
| "accuracy": 0.6355555555555555, | |
| "confusion_matrix": { | |
| "NOT_ENOUGH_INFO": { | |
| "NOT_ENOUGH_INFO": 61, | |
| "REFUTES": 12, | |
| "SUPPORTS": 39 | |
| }, | |
| "REFUTES": { | |
| "NOT_ENOUGH_INFO": 48, | |
| "REFUTES": 53, | |
| "SUPPORTS": 21 | |
| }, | |
| "SUPPORTS": { | |
| "NOT_ENOUGH_INFO": 29, | |
| "REFUTES": 15, | |
| "SUPPORTS": 172 | |
| } | |
| }, | |
| "examples": 450, | |
| "gold_label_distribution": { | |
| "NOT_ENOUGH_INFO": 112, | |
| "REFUTES": 122, | |
| "SUPPORTS": 216 | |
| }, | |
| "macro_f1": 0.5935365393682225, | |
| "per_class": { | |
| "NOT_ENOUGH_INFO": { | |
| "f1-score": 0.488, | |
| "precision": 0.4420289855072464, | |
| "recall": 0.5446428571428571, | |
| "support": 112.0 | |
| }, | |
| "REFUTES": { | |
| "f1-score": 0.5247524752475248, | |
| "precision": 0.6625, | |
| "recall": 0.4344262295081967, | |
| "support": 122.0 | |
| }, | |
| "SUPPORTS": { | |
| "f1-score": 0.7678571428571429, | |
| "precision": 0.7413793103448276, | |
| "recall": 0.7962962962962963, | |
| "support": 216.0 | |
| }, | |
| "accuracy": 0.6355555555555555, | |
| "macro avg": { | |
| "f1-score": 0.5935365393682225, | |
| "precision": 0.6153027652840247, | |
| "recall": 0.5917884609824501, | |
| "support": 450.0 | |
| }, | |
| "weighted avg": { | |
| "f1-score": 0.6322954329718686, | |
| "precision": 0.6454892831362097, | |
| "recall": 0.6355555555555555, | |
| "support": 450.0 | |
| } | |
| }, | |
| "predicted_label_distribution": { | |
| "NOT_ENOUGH_INFO": 138, | |
| "REFUTES": 80, | |
| "SUPPORTS": 232 | |
| }, | |
| "strategy": "top5_decisive", | |
| "supported_refuted_accuracy_when_evidence_retrieved_at_5": 0.7660377358490567, | |
| "supported_refuted_examples_with_evidence_retrieved_at_5": 265 | |
| } | |
| }, | |
| "retrieval": { | |
| "evaluated_queries": 450.0, | |
| "mrr@5": 0.5610740740740741, | |
| "ndcg@1": 0.4866666666666667, | |
| "ndcg@3": 0.5487532554897722, | |
| "ndcg@5": 0.5739882576045233, | |
| "recall@1": 0.4866666666666667, | |
| "recall@3": 0.6355555555555555, | |
| "recall@5": 0.6755555555555556 | |
| } | |
| }, | |
| "hub_model_id": "rishhh/verisci-claim-verifier-dense-adapted-seed123", | |
| "latency_seconds": 193.2834756150114, | |
| "limitations": [ | |
| "The derived test split is small and originates from SciFact validation, not an official hidden test set.", | |
| "Retrieval-aware negatives are generated by BM25 over SciFact train and may overfit SciFact retrieval artifacts.", | |
| "Full SciFact validation is public and should be treated as a promotion gate, not a hidden benchmark.", | |
| "SciFact is CC-BY-NC-2.0, so downstream use should respect non-commercial licensing constraints." | |
| ], | |
| "max_length": 512, | |
| "model_name": "rishhh/verisci-claim-verifier-retrieval-adapted-seed123", | |
| "seed": 123, | |
| "test": { | |
| "accuracy": 0.893491124260355, | |
| "confusion_matrix": { | |
| "NOT_ENOUGH_INFO": { | |
| "NOT_ENOUGH_INFO": 70, | |
| "REFUTES": 0, | |
| "SUPPORTS": 1 | |
| }, | |
| "REFUTES": { | |
| "NOT_ENOUGH_INFO": 4, | |
| "REFUTES": 20, | |
| "SUPPORTS": 7 | |
| }, | |
| "SUPPORTS": { | |
| "NOT_ENOUGH_INFO": 1, | |
| "REFUTES": 5, | |
| "SUPPORTS": 61 | |
| } | |
| }, | |
| "examples": 169, | |
| "expected_calibration_error": 0.09891798686699052, | |
| "gold_label_distribution": { | |
| "NOT_ENOUGH_INFO": 71, | |
| "REFUTES": 31, | |
| "SUPPORTS": 67 | |
| }, | |
| "macro_f1": 0.8567495491347223, | |
| "per_class": { | |
| "NOT_ENOUGH_INFO": { | |
| "f1-score": 0.958904109589041, | |
| "precision": 0.9333333333333333, | |
| "recall": 0.9859154929577465, | |
| "support": 71.0 | |
| }, | |
| "REFUTES": { | |
| "f1-score": 0.7142857142857143, | |
| "precision": 0.8, | |
| "recall": 0.6451612903225806, | |
| "support": 31.0 | |
| }, | |
| "SUPPORTS": { | |
| "f1-score": 0.8970588235294118, | |
| "precision": 0.8840579710144928, | |
| "recall": 0.9104477611940298, | |
| "support": 67.0 | |
| }, | |
| "accuracy": 0.893491124260355, | |
| "macro avg": { | |
| "f1-score": 0.8567495491347223, | |
| "precision": 0.8724637681159421, | |
| "recall": 0.847174848158119, | |
| "support": 169.0 | |
| }, | |
| "weighted avg": { | |
| "f1-score": 0.8895147343204121, | |
| "precision": 0.8893405368321755, | |
| "recall": 0.893491124260355, | |
| "support": 169.0 | |
| } | |
| }, | |
| "predicted_label_distribution": { | |
| "NOT_ENOUGH_INFO": 75, | |
| "REFUTES": 25, | |
| "SUPPORTS": 69 | |
| }, | |
| "strategy": "test" | |
| }, | |
| "training": { | |
| "base_scifact_entailment_train_examples": 916, | |
| "class_weights": { | |
| "NOT_ENOUGH_INFO": 0.9040744304656982, | |
| "REFUTES": 1.4886423349380493, | |
| "SUPPORTS": 0.8182346224784851 | |
| }, | |
| "dense_retrieved_nei_max": 350, | |
| "dense_retriever_id": "rishhh/verisci-scifact-e5-retriever", | |
| "effective_batch_size": 16, | |
| "epochs": 2, | |
| "hard_title_nei_max": 150, | |
| "hybrid_alpha": 0.75, | |
| "learning_rate": 6e-06, | |
| "retrieval_aware_examples_added": 2514, | |
| "retrieved_nei_max": 250, | |
| "test_examples": 169, | |
| "train_examples": 2818, | |
| "train_label_distribution": { | |
| "NOT_ENOUGH_INFO": 1039, | |
| "REFUTES": 631, | |
| "SUPPORTS": 1148 | |
| }, | |
| "validation_examples": 168, | |
| "validation_rows_removed_for_train_claim_overlap": 2 | |
| }, | |
| "validation": { | |
| "accuracy": 0.8630952380952381, | |
| "confusion_matrix": { | |
| "NOT_ENOUGH_INFO": { | |
| "NOT_ENOUGH_INFO": 55, | |
| "REFUTES": 0, | |
| "SUPPORTS": 2 | |
| }, | |
| "REFUTES": { | |
| "NOT_ENOUGH_INFO": 5, | |
| "REFUTES": 28, | |
| "SUPPORTS": 7 | |
| }, | |
| "SUPPORTS": { | |
| "NOT_ENOUGH_INFO": 3, | |
| "REFUTES": 6, | |
| "SUPPORTS": 62 | |
| } | |
| }, | |
| "examples": 168, | |
| "expected_calibration_error": 0.11418107826085311, | |
| "gold_label_distribution": { | |
| "NOT_ENOUGH_INFO": 57, | |
| "REFUTES": 40, | |
| "SUPPORTS": 71 | |
| }, | |
| "macro_f1": 0.8488876200143807, | |
| "per_class": { | |
| "NOT_ENOUGH_INFO": { | |
| "f1-score": 0.9166666666666666, | |
| "precision": 0.873015873015873, | |
| "recall": 0.9649122807017544, | |
| "support": 57.0 | |
| }, | |
| "REFUTES": { | |
| "f1-score": 0.7567567567567568, | |
| "precision": 0.8235294117647058, | |
| "recall": 0.7, | |
| "support": 40.0 | |
| }, | |
| "SUPPORTS": { | |
| "f1-score": 0.8732394366197183, | |
| "precision": 0.8732394366197183, | |
| "recall": 0.8732394366197183, | |
| "support": 71.0 | |
| }, | |
| "accuracy": 0.8630952380952381, | |
| "macro avg": { | |
| "f1-score": 0.8488876200143807, | |
| "precision": 0.8565949071334323, | |
| "recall": 0.8460505724404909, | |
| "support": 168.0 | |
| }, | |
| "weighted avg": { | |
| "f1-score": 0.860239703989704, | |
| "precision": 0.8613278644791249, | |
| "recall": 0.8630952380952381, | |
| "support": 168.0 | |
| } | |
| }, | |
| "predicted_label_distribution": { | |
| "NOT_ENOUGH_INFO": 63, | |
| "REFUTES": 34, | |
| "SUPPORTS": 71 | |
| }, | |
| "strategy": "validation" | |
| } | |
| } |