Create README.md

README.md

@@ -0,0 +1,76 @@
+---
+license: mit
+language:
+- id
+tags:
+- nlp
+- pytorch
+- xlstm
+- language-modeling
+- aspect-based-sentiment-analysis
+- sequence-labeling
+- indonesian
+pipeline_tag: token-classification
+---
+
+# Bi-xLSTM[7:1] for Indonesian End-to-End ABSA
+
+This repository contains a Bi-xLSTM[7:1] model pretrained on large-scale Indonesian Wikipedia data and fine-tuned for Indonesian End-to-End Aspect-Based Sentiment Analysis (E2E-ABSA).
+
+## Model Description
+
+The model uses a bidirectional xLSTM-based architecture for contextual language modeling. It was first pretrained on Indonesian Wikipedia data using a forward and backward language modeling objective, then fine-tuned for E2E-ABSA using BIOES sentiment tagging and CRF decoding.
+
+The final task is to extract aspect–sentiment pairs directly from Indonesian review text.
+
+## Architecture
+
+- Model: Bi-xLSTM[7:1]
+- Pretraining objective: Bidirectional contextual language modeling
+- Fine-tuning task: End-to-End Aspect-Based Sentiment Analysis
+- Decoder: CRF
+- Labeling scheme: BIOES with sentiment labels
+- Framework: PyTorch
+- Language: Indonesian
+
+## Dataset
+
+The model was pretrained using Indonesian Wikipedia data and fine-tuned on Indonesian review data for aspect-based sentiment analysis.
+
+## Intended Use
+
+This model is intended for research and academic purposes, especially for:
+
+- Indonesian NLP
+- Sequence labeling
+- Aspect-Based Sentiment Analysis
+- Contextual language modeling
+- Comparison between xLSTM-based models and Transformer-based models
+
+## Files
+
+Recommended repository files:
+
+- `checkpoint.pt` or `pytorch_model.bin`: trained model checkpoint
+- `model.py`: Bi-xLSTM[7:1] model architecture
+- `config.json`: model configuration
+- `label2id.json`: label-to-index mapping
+- `id2label.json`: index-to-label mapping
+- `requirements.txt`: required Python libraries
+- `inference_example.py`: example inference script
+
+## Usage
+
+This model uses a custom PyTorch architecture, so it is not directly loadable using `AutoModel.from_pretrained()` unless a custom Hugging Face Transformers wrapper is added.
+
+Example loading format:
+
+```python
+import torch
+from model import BiXLSTMCRF
+
+checkpoint = torch.load("checkpoint.pt", map_location="cpu")
+
+model = BiXLSTMCRF(**checkpoint["config"])
+model.load_state_dict(checkpoint["model_state_dict"])
+model.eval()