Hugging Face's logo

TomBombadyl
/
Qwen2.5-Coder-7B-Instruct-Omni1.1

Text Generation
Transformers
Safetensors
GGUF
English
robotics
isaac-sim
code-generation
simulation
qwen2
causal-lm
text2text-generation
omni
nvidia
robotics-simulation
Model card Files
xet
Community
Qwen2.5-Coder-7B-Instruct-Omni1.1 / model_card.md
TomBombadyl's picture
TomBombadyl
Upload model_card.md with huggingface_hub
29f9bcb verified
preview code
|
raw
history blame contribute delete
7.54 kB

Model Card for Isaac Sim Robotics Qwen2.5-Coder-7B-Instruct

Model Details

Model Description

  • Model Type: Fine-tuned causal language model
  • Base Model: Qwen/Qwen2.5-Coder-7B-Instruct
  • Architecture: Qwen2 architecture with 7B parameters
  • Training Method: LoRA (Low-Rank Adaptation) fine-tuning
  • License: MIT License
  • Repository: Qwen2.5-Coder-7B-Instruct-Omni1.1

Intended Use

This model is specifically designed for Isaac Sim 5.0 robotics development tasks, including:

  • Robot simulation setup and configuration
  • Computer vision and sensor integration
  • Robot control programming
  • Simulation environment design
  • Troubleshooting Isaac Sim issues
  • Code generation for robotics workflows

Training Data

  • Source: Isaac Sim 5.0 Synthetic Dataset
  • Total Samples: 2,000 carefully curated examples
  • Training Split: 1,800 training, 200 evaluation
  • Data Types:
    • Robot creation and configuration
    • Sensor setup and data processing
    • Physics parameter tuning
    • Environment design
    • Troubleshooting scenarios
  • Curriculum Learning: Applied (sorted by output length)

Training Configuration

  • LoRA Rank: 64
  • LoRA Alpha: 128
  • Learning Rate: 1e-05
  • Batch Size: 1
  • Gradient Accumulation Steps: 8
  • Max Training Steps: 300
  • Warmup Steps Ratio: 0.03
  • Optimizer: AdamW
  • Scheduler: Linear with warmup

Hardware Requirements

  • Training GPU: NVIDIA GeForce RTX 4070 Laptop GPU
  • VRAM: 8.5GB
  • Inference Requirements:
    • HuggingFace: 8GB+ VRAM (full precision)
    • CTransformers: 4GB+ VRAM (optimized)
    • GGUF: 2GB+ VRAM (when conversion is fixed)

Performance

Evaluation Metrics

  • Training Loss: Converged after 300 steps
  • Domain Accuracy: Specialized for Isaac Sim robotics
  • Code Quality: Generated code follows Isaac Sim best practices
  • Response Relevance: High relevance to robotics queries

Limitations

  1. Domain Specificity: Limited to Isaac Sim robotics context
  2. GGUF Conversion: Currently has metadata compatibility issues
  3. Hardware Requirements: Requires significant VRAM for full precision
  4. Training Data Size: Limited to 2,000 examples

Known Issues

  • GGUF Loading Error: Missing qwen2.context_length metadata field
  • Workaround: Use HuggingFace or CTransformers formats
  • Status: Under investigation for future updates

Usage

Input Format

The model expects Isaac Sim-specific queries in the following format:

<|im_start|>user
[Your Isaac Sim robotics question here]
<|im_end|>
<|im_start|>assistant

Example Queries

  1. Robot Creation: "How do I create a differential drive robot in Isaac Sim?"
  2. Sensor Setup: "How to add a depth camera and process depth data?"
  3. Physics Configuration: "What physics parameters should I use for a manipulator?"
  4. Environment Design: "How to create a warehouse environment with obstacles?"
  5. Troubleshooting: "Why is my robot falling through the ground?"

Output Characteristics

  • Code Generation: Python scripts ready for Isaac Sim
  • Explanation: Detailed step-by-step instructions
  • Best Practices: Follows Isaac Sim development guidelines
  • Error Prevention: Includes common pitfalls and solutions

Model Variants

1. HuggingFace Format (Primary)

  • Location: models/huggingface/
  • Size: 5.3GB
  • Format: Standard HuggingFace model files
  • Usage: Direct integration with transformers library
  • Advantages: Full compatibility, easy integration

2. CTransformers Format (Alternative)

  • Location: models/ctransformers/
  • Size: 5.2GB
  • Format: Optimized for CTransformers library
  • Usage: Lightweight inference with reduced memory
  • Advantages: Lower memory usage, faster inference

3. GGUF Format (Experimental)

  • Location: models/gguf/
  • Size: 616MB (base) + quantization variants
  • Format: llama.cpp compatible
  • Usage: Server deployment and edge inference
  • Status: Metadata issues, conversion scripts provided

Ethical Considerations

Bias and Fairness

  • Training Data: Focused on technical robotics content
  • Domain Limitation: May not generalize to other robotics platforms
  • Cultural Bias: Minimal, focused on technical accuracy

Safety

  • Content Filtering: No additional safety filters applied
  • Use Case: Intended for robotics development only
  • Misuse Prevention: Technical domain limits potential misuse

Privacy

  • Training Data: Synthetic data, no personal information
  • Inference: No data collection or logging
  • Compliance: Follows standard AI model privacy practices

Technical Specifications

Model Architecture

  • Base: Qwen2.5-Coder-7B-Instruct
  • Parameters: 7 billion
  • Context Length: 32,768 tokens
  • Vocabulary: 151,936 tokens
  • Embedding Dimension: 4,096
  • Attention Heads: 32
  • Layers: 32

Quantization Support

  • FP16: Full precision (default)
  • INT8: 8-bit quantization support
  • INT4: 4-bit quantization (experimental)
  • GGUF: Conversion scripts provided

Integration

  • HuggingFace: Native support
  • Isaac Sim: Direct Python integration
  • CTransformers: Optimized inference
  • llama.cpp: When GGUF issues resolved

Deployment

Local Development 

# Clone repository
git clone https://github.com/your-username/isaac-sim-robotics-qwen.git
cd isaac-sim-robotics-qwen

# Install dependencies
pip install -r requirements.txt

# Download model
huggingface-cli download your-username/isaac-sim-robotics-qwen

Production Deployment

  • HuggingFace Hub: Direct model hosting
  • Docker: Containerized deployment
  • API Server: RESTful inference endpoints
  • Edge Deployment: GGUF format (when fixed)

Maintenance

Updates

  • Training Data: Expandable dataset for future versions
  • Model Architecture: Base model updates as available
  • Bug Fixes: Regular repository updates
  • Community: Open source maintenance

Support

  • Documentation: Comprehensive guides and examples
  • Issues: GitHub issue tracking
  • Discussions: Community support forum
  • Examples: Working code samples

Citation

If you use this model in your research or development, please cite:

@misc{qwen2.5_coder_7b_instruct_omni1.1,
  title={Qwen2.5-Coder-7B-Instruct-Omni1.1: Isaac Sim Robotics Specialized Model},
  author={TomBombadyl},
  year={2025},
  url={https://huggingface.co/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.1}
}

License

This model is licensed under the MIT License. See the LICENSE file for details.

Contact

Note: This model is specifically trained for Isaac Sim 5.0 robotics development. For general coding tasks, consider using the base Qwen2.5-Coder-7B-Instruct model.