README.md

---
license: apache-2.0
base_model: Qwen/Qwen2.5-0.5B-Instruct
library_name: transformers
language:
- en
pipeline_tag: text-generation
tags:
- argument-mining
- fact-checking
- information-extraction
- qwen2
- peft
- lora
- research
---

# ArgParser-v1-Qwen0.5B

> **Version 1 of the ArgParser model series** — the initial baseline developed for structured argument extraction using a compact Qwen model.

ArgParser-v1-Qwen0.5B is the first model in the **ArgParser** research series. It serves as the baseline for evaluating lightweight argument-structure extraction models and demonstrates the feasibility of extracting structured argumentative information using a small language model.

This version performs a full fine-tuning of **Qwen2.5-0.5B-Instruct** on four academic argument-mining datasets and establishes the foundation for the later LoRA-based models in the series.

---

# Model Overview

| Property | Value |
|----------|-------|
| Model | ArgParser-v1-Qwen0.5B |
| Base Model | Qwen/Qwen2.5-0.5B-Instruct |
| Task | Argument Structure Extraction |
| Training Method | Full Fine-tuning |
| Author | **Jayesh Choudhari** |
| License | Apache-2.0 |

---

# Training Configuration

This model was trained using four manually annotated academic argument-mining corpora:

- AbstRCT
- Microtext
- CDCP
- PERSPECTRUM

Training configuration:

- **Training Samples:** 1,494
- **Epochs:** 3
- **Optimizer:** Adafactor
- **Precision:** FP16
- **Hardware:** NVIDIA GTX 1080 Ti
- **Training Time:** ~1.5 hours

---

# Performance

Average held-out Component F1 across the four datasets:

**Component F1:** **0.108**

Highlights:

- Best performance:
  - CDCP Claim Extraction → **0.501 F1**
- Weakest performance:
  - PERSPECTRUM
  - Approximately **91% empty predictions**

As expected, this early baseline struggles to generalize beyond academic argument-mining datasets.

---

# Intended Task

The model extracts structured argumentative information from input text, including:

- Claims
- Premises
- Citations
- Argument relations

The generated output follows a structured JSON format suitable for downstream NLP pipelines.

---

# Usage

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

base_id = "Qwen/Qwen2.5-0.5B-Instruct"
adapter_id = "iamjayeshc/ArgParser-v1-Qwen0.5B"

tokenizer = AutoTokenizer.from_pretrained(base_id)

base = AutoModelForCausalLM.from_pretrained(
    base_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

model = PeftModel.from_pretrained(base, adapter_id)
```

Example prompt:

```python
instruction = (
    "Extract all argument components and relations from the text. "
    "Return strict JSON containing "
    "claim_components, premise_components, citation_components and relations."
)

text = "The Obama administration is putting Border Patrol agents in a chokehold."
```

---

# Model Series

ArgParser-v1 is the first model in the development series.

If you're looking for improved performance, consider one of the later versions:

- **ArgParser-v2-Qwen1.5B**
  - https://huggingface.co/iamjayeshc/ArgParser-v2-Qwen1.5B

- **ArgParser-v3-Qwen1.5B**
  - https://huggingface.co/iamjayeshc/ArgParser-v3-Qwen1.5B

- **ArgParser-v4-Qwen1.5B (Recommended)**
  - https://huggingface.co/iamjayeshc/ArgParser-v4-Qwen1.5B

ArgParser-v4 is the final distilled model in this series and provides substantially better downstream performance on the target fact-checking task.

---

# Limitations

This model represents the initial baseline and has several limitations:

- Trained only on academic argument-mining corpora.
- Limited transfer to informal language such as social media or political fact-checking claims.
- Lower extraction accuracy compared with later versions.

It is primarily provided for **research reproducibility** and comparison with subsequent iterations.

---

# Acknowledgements

This model is part of the **ArgParser** research project investigating lightweight argument-structure extraction through progressively improved model distillation and domain adaptation.

Released by **Jayesh Choudhari**.

---

# Citation

If you use this model in your research, please cite the associated project when available.

A formal technical report / preprint will be released in the future.