README.md

---
license: mit
language:
  - en
  - code
tags:
  - type-inference
  - typescript
  - code-generation
  - type-ground
  - peft
  - lora
  - code-t5
  - unixcoder
  - llama
  - qwen
  - deepseek
pipeline_tag: text-generation
datasets:
  - TypeGround
  - ManyTypes4TypeScript
---

# TypeGround_weight

Model weights for the paper **"TypeGround: Fine-Grained Benchmarking for TypeScript Type Inference"**.

> [TypeGround](https://github.com/fumx66/TypeGround)

## Usage

### Traditional Models (Full Fine-tune)

```python
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model = AutoModelForSeq2SeqLM.from_pretrained("./CodeT5/TypeGround")
tokenizer = AutoTokenizer.from_pretrained("./CodeT5/TypeGround")
```

### LLMs (LoRA Adapters)

| Directory | Base Model |
|---|---|
| `Llama3-8B` | `meta-llama/Meta-Llama-3-8B-Instruct` |
| `Qwen3-14B` | `Qwen/Qwen3-14B` |
| `DeepSeek-Coder-6.7B` | `deepseek-ai/deepseek-coder-6.7b-instruct` |

```bash
pip install vllm
```

```bash
vllm serve meta-llama/Meta-Llama-3-8B-Instruct \
    --enable-lora \
    --lora-modules my-lora=./Llama3-8B/ManyTypes4TypeScrip/lora/sft \
    --max-lora-rank 8
```

### Batch Prediction

```bash
python prediction.py
```

## Models

| Model | Architecture | Type | LoRA Config |
|---|---|---|---|
| CodeT5 | T5ForConditionalGeneration | Full fine-tune | — |
| CodeT5+ | T5ForConditionalGeneration | Full fine-tune | — |
| UniXcoder | UniXcoder | Full fine-tune | — |
| Llama3-8B | CausalLM + LoRA | Adapter | rank=8, α=16 |
| Qwen3-14B | CausalLM + LoRA | Adapter | rank=8, α=16 |
| DeepSeek-Coder-6.7B | CausalLM + LoRA | Adapter | rank=8, α=16 |

## Citation

```bibtex
@inproceedings{typeground,
  title     = {TypeGround: Fine-Grained Benchmarking for TypeScript Type Inference},
  author    = {Anonymous},
  booktitle = {},
  year      = {2026},
  url       = {https://github.com/fumx66/TypeGround}
}
```

## License

MIT License