---
language: en
license: mit
tags:
  - log-analysis
  - anomaly-detection
  - embeddings
  - sentence-transformers
  - bge
base_model: BAAI/bge-large-en-v1.5
pipeline_tag: sentence-similarity
---

# BGE Log Embedding Model

Fine-tuned [`BAAI/bge-large-en-v1.5`](https://huggingface.co/BAAI/bge-large-en-v1.5) for **log anomaly detection**.

## Usage
```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("Swapnanil09/bge-log-embeddings")
logs = ["INFO: login ok", "ERROR: DB timeout"]
embeddings = model.encode(logs, normalize_embeddings=True)
print(embeddings.shape)  # (2, 1024)
```

## Training Details
| Property | Value |
|---|---|
| Base Model | BAAI/bge-large-en-v1.5 |
| Embedding Dim | 1024 |
| Loss | TripletLoss |
| Epochs | 3 |
| Batch Size | 32 |
| Hardware | Kaggle T4 x2 |

## How It Works
Normal logs cluster together; ERROR/CRITICAL anomalies are pushed apart via triplet loss.
Use the embeddings with DBSCAN or KMeans for zero-shot anomaly detection.