README.md

---
license: mit
datasets:
- mhhmm/typescript-instruct-20k
base_model:
- google/gemma-4-12B
tags:
- typescript
- finetuned
- spaceoutpl
- unsloth
- code
- text-generation
- instruct
- programming
- gemma
- lora
- javascript
---

# Gemma-4-12B TypeScript

This model is a specialized, fine-tuned version of [Google's Gemma-4 12B](https://huggingface.co/google/gemma-4-12B) designed specifically for TypeScript code generation, refactoring, and instruction. It was fine-tuned efficiently using the [Unsloth](https://github.com/unslothai/unsloth) library by [spaceoutpl].

## 💻 Model Details

* **Base Model:** google/gemma-4-12B
* **License:** MIT
* **Language:** English / TypeScript
* **Fine-tuning Framework:** Unsloth
* **Dataset:** [mhhmm/typescript-instruct-20k](https://huggingface.co/datasets/mhhmm/typescript-instruct-20k)

## 🚀 Intended Use

This model is ideal for developers and researchers looking for an AI assistant heavily specialized in the TypeScript ecosystem. Use cases include:
* **Code Generation:** Writing complex TypeScript functions, interfaces, and types.
* **Code Refactoring:** Converting standard JavaScript to strictly-typed TypeScript.
* **Instruction & Explanation:** Explaining TypeScript errors, generics, utility types, and best practices.

## 🛠️ Getting Started

You can load this model directly using the `transformers` library. Since it was trained with Unsloth, you can also utilize Unsloth's optimized inference engines for faster generation.

### Installation

```bash
pip install transformers torch accelerate
```

### Usage (Hugging Face Transformers)

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

model_id = "spaceoutpl/gemma-4-12B-typescript" # Update with your actual repo name

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.bfloat16
)

prompt = "Write a generic TypeScript function to fetch and strictly type data from an API."
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## 📊 Training Data

The model was fine-tuned on the `mhhmm/typescript-instruct-20k` dataset, which contains 20,000 high-quality instructional pairs focusing on TypeScript programming concepts, syntax, and problem-solving.

## ⚠️ Limitations & Biases

* **Hallucinations:** Like all LLMs, the model may occasionally generate plausible-looking but syntactically incorrect or non-compiling TypeScript code. Always test generated code in your environment.
* **Knowledge Cutoff:** The model's knowledge is limited to the training data of the base Gemma-4 model and the specific TypeScript dataset used for fine-tuning. It may not reflect the absolute latest TypeScript beta features.