sohomn/log-ner-deberta-lora

DeBERTa-v3 Log Entity Recognition Model

Fine-tuned DeBERTa-v3-small for Named Entity Recognition on system and cloud logs.

Model Details

• Base Model: microsoft/deberta-v3-base
• Training Data: 7003 synthetic + real logs
• Validation F1: Check evaluation_results.txt

Entities

['O', 'B-SERVICE', 'I-SERVICE', 'B-ERROR', 'I-ERROR', 'B-HOST', 'I-HOST', 'B-PROCESS', 'I-PROCESS']

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification
from peft import PeftModel

model_id = "YOUR_USERNAME/log-ner-deberta-lora"

tokenizer = AutoTokenizer.from_pretrained(model_id)
base_model = AutoModelForTokenClassification.from_pretrained("microsoft/deberta-v3-base")
model = PeftModel.from_pretrained(base_model, model_id)

# Extract entities
text = "nginx timeout on server1"
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)

Training Configuration

• LoRA rank: 32
• Training epochs: 15
• Learning rate: 0.0003