|
|
--- |
|
|
language: en |
|
|
tags: |
|
|
- qwen2 |
|
|
- lora |
|
|
- fine-tuned |
|
|
- code-generation |
|
|
- opencodeinstruct |
|
|
license: apache-2.0 |
|
|
base_model: Qwen/Qwen2-0.5B-Instruct |
|
|
--- |
|
|
|
|
|
# Qwen2-0.5b LoRA Fine-tuned on OpenCodeInstruct |
|
|
|
|
|
This model is a LoRA fine-tuned version of Qwen/Qwen2-0.5B-Instruct on the OpenCodeInstruct dataset. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Base Model**: Qwen/Qwen2-0.5B-Instruct |
|
|
- **Fine-tuning Dataset**: OpenCodeInstruct (300 samples) |
|
|
- **Fine-tuning Method**: LoRA (Low-Rank Adaptation) |
|
|
- **LoRA Rank**: 16 |
|
|
- **LoRA Alpha**: 32 |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
from peft import PeftModel |
|
|
|
|
|
# Load base model |
|
|
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-0.5B-Instruct") |
|
|
|
|
|
# Load LoRA adapters |
|
|
model = PeftModel.from_pretrained(base_model, "alpayH/qwen2-0.5b-lora-opencodeinstruct") |
|
|
|
|
|
# Load tokenizer |
|
|
tokenizer = AutoTokenizer.from_pretrained("alpayH/qwen2-0.5b-lora-opencodeinstruct") |
|
|
|
|
|
# Generate code |
|
|
prompt = "### Instruction:\nWrite a Python function to reverse a string\n\n### Response:\n" |
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
outputs = model.generate(**inputs, max_length=512) |
|
|
print(tokenizer.decode(outputs[0])) |
|
|
``` |
|
|
|
|
|
## Training Details |
|
|
|
|
|
- **Learning Rate**: 2e-4 |
|
|
- **Batch Size**: 16 (effective, with gradient accumulation) |
|
|
- **Epochs**: 3 |
|
|
- **Precision**: bfloat16 |
|
|
|
|
|
## Evaluation |
|
|
|
|
|
This model has been evaluated on LiveCodeBench. See the main repository for evaluation results. |
|
|
|
|
|
## License |
|
|
|
|
|
Apache 2.0 |
|
|
|
