Internal-LinkLM

A fine-tuned Cross-Encoder Reranker based on BAAI/bge-reranker-large for semantic document ranking and internal link recommendation.

This model was fine-tuned to score the relevance between a query and a candidate document. It is designed for retrieval and reranking tasks such as:

• Internal Link Recommendation
• Semantic Search
• Information Retrieval
• Document Ranking
• Search Result Reranking

---

Base Model

• BAAI/bge-reranker-large
• Architecture: XLM-RoBERTa Cross Encoder

---

Training

The model was fine-tuned using the Hugging Face Transformers Trainer.

Training configuration:

• Epochs: 3
• Framework: PyTorch
• Library: Transformers
• Loss: Cross-Entropy Loss
• Optimizer: AdamW

---

Evaluation

Metric	Score
Accuracy	0.8476
Precision	0.6279
Recall	0.5841
F1 Score	0.6052

---

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained(
    "NayonAhmed09/internal-linklm"
)

model = AutoModelForSequenceClassification.from_pretrained(
    "NayonAhmed09/internal-linklm"
)

Example:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained(
    "NayonAhmed09/internal-linklm"
)

model = AutoModelForSequenceClassification.from_pretrained(
    "NayonAhmed09/internal-linklm"
)

query = "Best moisturizer for oily skin"
document = "A complete guide to choosing skincare products for oily skin."

inputs = tokenizer(
    query,
    document,
    return_tensors="pt",
    truncation=True,
    max_length=512
)

with torch.no_grad():
    score = model(**inputs).logits.squeeze().item()

print(score)

Higher scores indicate higher semantic relevance.

---

Files

• model.safetensors
• config.json
• tokenizer.json
• tokenizer_config.json
• training_args.bin

---

License

MIT License

---

Author

Nayon Ahmed

GitHub: https://github.com/nayonahmedjoy

Hugging Face: https://huggingface.co/NayonAhmed09