metadata
license: llama3.2
language: en
library_name: transformers
pipeline_tag: text-generation
tags:
- unreal-engine
- game-development
- code
- fine-tuned
- nlp
base_model: meta-llama/CodeLlama-7b-hf
extra_gated_prompt: |
⚠️ This model is proprietary and access is restricted.
Downloading requires explicit approval from the model owner.
Please fill out the form below and wait for approval.
extra_gated_fields:
Name: text
Company/Organization: text
Email: text
Intended Use Case: text
I agree to use this model only for authorized purposes: checkbox
Parser-CodeLlama-7B v0
A fine-tuned CodeLlama-7B model that converts natural language commands into structured Unreal Engine JSON property modifications.
Model Description
This model was trained to interpret natural language commands in the context of Unreal Engine development and output structured JSON that can be directly used to modify object properties.
Capabilities
- Movement commands: "Move the box forward 100 units"
- Rotation commands: "Rotate 90 degrees to the right"
- Scale commands: "Make this twice as big"
- Spawn commands: "Create a sphere here"
- Delete commands: "Remove this object"
Performance
| Metric | Score |
|---|---|
| Structure Accuracy | 100% |
| Value Accuracy | ~90% |
| JSON Validity | 100% |
Example
Input:
User Command: "Rotate the cube 90 degrees"
Object State: "Cube"
{"RelativeRotation": "(Pitch=0.0,Yaw=45.0,Roll=0.0)"}
Output:
[{
"Action": "Edit",
"PropertyName": "RelativeRotation",
"PropertyType": "StructProperty",
"Value": "(Pitch=0.0,Yaw=135.0,Roll=0.0)"
}]
Technical Details
| Base Model | meta-llama/CodeLlama-7b-hf |
| Fine-tuning Method | LoRA |
| Training Examples | 5,000 |
| LoRA Rank | 16 |
| Training Iterations | 800 |
| Final Loss | 0.096 |
Requirements
| Platform | Library | Memory |
|---|---|---|
| Apple Silicon | mlx-lm |
~14GB |
| NVIDIA GPU | transformers + torch |
~14GB VRAM |
Important Notes
- System prompt required: This model requires a specific system prompt to achieve optimal performance. Contact the maintainers for documentation.
- Output processing: Model output may contain special tokens that need to be stripped before JSON parsing.
License
This model is subject to the Llama 3.2 Community License and additional usage restrictions.
Contact
For access requests or questions, please submit a request through the Hugging Face interface.