57 lines
2.3 KiB
Markdown
57 lines
2.3 KiB
Markdown
|
---
|
||
|
|
language: en
|
||
|
|
license: apache-2.0
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
tags:
|
||
|
|
- conversational
|
||
|
|
- code
|
||
|
|
datasets:
|
||
|
|
- smirki/Agentic-Coding-Tessa
|
||
|
|
library_name: transformers
|
||
|
|
---
|
||
|
|
# Nero1-0.5B
|
||
|
|
|
||
|
|
**Nero1-0.5B** is a specialized, lightweight coding model developed by **NeuronicL**. It is a **full fine-tune** of `Qwen/Qwen2.5-Coder-0.5B-Instruct`, specifically optimized for agentic workflows, tool use, and complex code generation tasks using the `smirki/Agentic-Coding-Tessa` dataset.
|
||
|
|
|
||
|
|
## Model Description
|
||
|
|
|
||
|
|
Unlike standard parameter-efficient fine-tuning (LoRA), Nero1-0.5B underwent a full parameter update. This allows the model to deeply integrate the agentic reasoning patterns found in the Tessa dataset, making it exceptionally capable of:
|
||
|
|
- Writing functional, production-ready code.
|
||
|
|
- Understanding and executing multi-step agentic instructions.
|
||
|
|
- Maintaining high performance in low-latency environments (Edge/Local development).
|
||
|
|
|
||
|
|
### Key Specifications
|
||
|
|
- **Base Model:** [Qwen/Qwen2.5-Coder-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct)
|
||
|
|
- **Training Data:** [smirki/Agentic-Coding-Tessa](https://huggingface.co/datasets/smirki/Agentic-Coding-Tessa)
|
||
|
|
- **Fine-tuning Method:** Full Parameter Fine-tuning (Full FT)
|
||
|
|
- **Parameters:** 0.49 Billion
|
||
|
|
- **Context Length:** 32,768 tokens
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
|
||
|
|
You can use Nero1-0.5B with the Hugging Face `transformers` library. Given its Qwen2.5-Coder backbone, it follows the standard ChatML-style prompt template.
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
|
||
|
|
model_name = "NeuronicL/Nero1-0.5B"
|
||
|
|
device = "cuda" # or "cpu"
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
||
|
|
model_name,
|
||
|
|
torch_dtype="auto",
|
||
|
|
device_map="auto"
|
||
|
|
)
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||
|
|
|
||
|
|
messages = [
|
||
|
|
{"role": "system", "content": "You are a helpful coding assistant specialized in agentic tasks."},
|
||
|
|
{"role": "user", "content": "Write a Python script to scrape news headlines and save them to a JSON file."}
|
||
|
|
]
|
||
|
|
|
||
|
|
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
||
|
|
model_inputs = tokenizer([text], return_tensors="pt").to(device)
|
||
|
|
|
||
|
|
generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=512)
|
||
|
|
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
||
|
|
print(response)
|