Add pipeline tag and improve model card

This PR improves the model card by adding the `text-generation` pipeline tag to the metadata, which ensures the model is correctly categorized on the Hugging Face Hub. It also updates the README with key features of the AnalogToBi framework, usage instructions, and a BibTeX citation based on the associated research paper and GitHub repository.

README.md

@@ -1,7 +1,8 @@
 ---
-license: mit
 language:
 - en
+license: mit
+pipeline_tag: text-generation
 tags:
 - analog-circuits
 - circuit-generation
@@ -11,35 +12,48 @@ tags:
 
 # AnalogToBi
 
-AnalogToBi is a generative framework for device-level analog circuit topology generation.  
-The model generates valid analog circuit topologies conditioned on a target circuit type using a Transformer decoder.
+AnalogToBi is a generative framework for device-level analog circuit topology generation, introduced in the paper [AnalogToBi: Device-Level Analog Circuit Topology Generation via Bipartite Graph and Grammar Guided Decoding](https://huggingface.co/papers/2603.08720). 
 
-Key ideas of AnalogToBi include:
+The model generates valid and novel analog circuit topologies conditioned on a target circuit type using a Transformer decoder.
 
-- Circuit-type token for explicit functional control
-- Bipartite graph circuit representation separating devices and nets
-- Grammar-guided decoding to enforce electrical validity
-- Device renaming data augmentation to improve generalization
+## Key Features
 
-Experimental results show that AnalogToBi achieves 97.8% validity and 92.1% novelty in generated circuits.
+- **Circuit-type conditioning**: Explicit functional control across 15 circuit categories (e.g., OpAmp, LDO, Comparator).
+- **Bipartite graph representation**: Decouples devices and nets into distinct node types for better structural generalization.
+- **Grammar-guided decoding**: State machine-based constrained decoding enforces electrical validity during generation.
+- **Device renaming augmentation**: Randomizes device numbering to mitigate memorization and improve novelty.
+
+Experimental results show that AnalogToBi achieves 97.8% validity and 92.1% novelty in generated circuits without human-in-the-loop training.
 
 ---
 
 ## Paper
 
-AnalogToBi: Device-Level Analog Circuit Topology Generation via Bipartite Graph and Grammar Guided Decoding
-
-arXiv: https://arxiv.org/abs/2603.08720
+[AnalogToBi: Device-Level Analog Circuit Topology Generation via Bipartite Graph and Grammar Guided Decoding](https://arxiv.org/abs/2603.08720)
 
 ---
 
 ## Code
 
-Official implementation:  
-https://github.com/Seungmin0825/AnalogToBi
+Official implementation: [https://github.com/Seungmin0825/AnalogToBi](https://github.com/Seungmin0825/AnalogToBi)
 
 ---
 
-## Model Weights
+## Usage
+
+To generate circuit topologies using the grammar-guided decoder, you can use the following command from the official repository:
+
+```bash
+python GPT_Inference_Grammar.py CIRCUIT_Opamp
+```
+
+## Citation
 
-This repository provides pretrained model checkpoints for AnalogToBi.
+```bibtex
+@article{kim2026analogtobi,
+  title={AnalogToBi: Device-Level Analog Circuit Topology Generation via Bipartite Graph and Grammar Guided Decoding},
+  author={Kim, Seungmin and Kim, Mingun and Lee, Yuna and Kim, Yulhwa},
+  journal={arXiv preprint arXiv:2603.08720},
+  year={2026}
+}
+```