Shadow-Coder v2 LoRA
Fine-tuned from Qwen/Qwen2.5-Coder-3B-Instruct on fullstack coding tasks across multiple languages and frameworks.
Model Details
| Property | Value |
|---|---|
| Base Model | Qwen2.5-Coder-3B-Instruct |
| Method | LoRA (r=16, alpha=32) |
| GPU | AMD Radeon RX 9060 XT (ROCm) |
| Framework | Unsloth + HuggingFace TRL |
| Epochs | 3 |
Training Data
- Custom fullstack dataset (750 examples)
- CodeAlpaca-20k (5,000 examples)
- Magicoder-OSS-Instruct-75K (3,000 examples)
- Total: ~8,750 examples
Languages & Frameworks
Python, JavaScript, TypeScript, Rust, Go, SQL, PHP, FastAPI, React, Vue, NestJS, Express, PostgreSQL, Docker
Usage
from unsloth import FastLanguageModel
import torch
model, tokenizer = FastLanguageModel.from_pretrained(
model_name = "lhordking/Shadow-coder-v2-LoRA",
max_seq_length = 2048,
dtype = torch.bfloat16,
load_in_4bit = False,
)
FastLanguageModel.for_inference(model)
prompt = (
"### Instruction:\n"
"Build a FastAPI endpoint for user authentication with JWT\n\n"
"### Context:\n"
"Use PostgreSQL and return access + refresh tokens\n\n"
"### Response:\n"
)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(
**inputs,
max_new_tokens = 512,
temperature = 0.7,
repetition_penalty = 1.3,
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Training Hardware
- GPU: AMD Radeon RX 9060 XT 16GB
- ROCm: 7.0
- OS: Ubuntu 24.04
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support