Upload model_card.md with huggingface_hub
Browse files- model_card.md +219 -0
model_card.md
ADDED
@@ -0,0 +1,219 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Model Card for Isaac Sim Robotics Qwen2.5-Coder-7B-Instruct
|
2 |
+
|
3 |
+
## Model Details
|
4 |
+
|
5 |
+
### Model Description
|
6 |
+
- **Model Type**: Fine-tuned causal language model
|
7 |
+
- **Base Model**: Qwen/Qwen2.5-Coder-7B-Instruct
|
8 |
+
- **Architecture**: Qwen2 architecture with 7B parameters
|
9 |
+
- **Training Method**: LoRA (Low-Rank Adaptation) fine-tuning
|
10 |
+
- **License**: MIT License
|
11 |
+
- **Repository**: [Qwen2.5-Coder-7B-Instruct-Omni1.1](https://huggingface.co/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.1)
|
12 |
+
|
13 |
+
### Intended Use
|
14 |
+
This model is specifically designed for Isaac Sim 5.0 robotics development tasks, including:
|
15 |
+
- Robot simulation setup and configuration
|
16 |
+
- Computer vision and sensor integration
|
17 |
+
- Robot control programming
|
18 |
+
- Simulation environment design
|
19 |
+
- Troubleshooting Isaac Sim issues
|
20 |
+
- Code generation for robotics workflows
|
21 |
+
|
22 |
+
### Training Data
|
23 |
+
- **Source**: Isaac Sim 5.0 Synthetic Dataset
|
24 |
+
- **Total Samples**: 2,000 carefully curated examples
|
25 |
+
- **Training Split**: 1,800 training, 200 evaluation
|
26 |
+
- **Data Types**:
|
27 |
+
- Robot creation and configuration
|
28 |
+
- Sensor setup and data processing
|
29 |
+
- Physics parameter tuning
|
30 |
+
- Environment design
|
31 |
+
- Troubleshooting scenarios
|
32 |
+
- **Curriculum Learning**: Applied (sorted by output length)
|
33 |
+
|
34 |
+
### Training Configuration
|
35 |
+
- **LoRA Rank**: 64
|
36 |
+
- **LoRA Alpha**: 128
|
37 |
+
- **Learning Rate**: 1e-05
|
38 |
+
- **Batch Size**: 1
|
39 |
+
- **Gradient Accumulation Steps**: 8
|
40 |
+
- **Max Training Steps**: 300
|
41 |
+
- **Warmup Steps Ratio**: 0.03
|
42 |
+
- **Optimizer**: AdamW
|
43 |
+
- **Scheduler**: Linear with warmup
|
44 |
+
|
45 |
+
### Hardware Requirements
|
46 |
+
- **Training GPU**: NVIDIA GeForce RTX 4070 Laptop GPU
|
47 |
+
- **VRAM**: 8.5GB
|
48 |
+
- **Inference Requirements**:
|
49 |
+
- **HuggingFace**: 8GB+ VRAM (full precision)
|
50 |
+
- **CTransformers**: 4GB+ VRAM (optimized)
|
51 |
+
- **GGUF**: 2GB+ VRAM (when conversion is fixed)
|
52 |
+
|
53 |
+
## Performance
|
54 |
+
|
55 |
+
### Evaluation Metrics
|
56 |
+
- **Training Loss**: Converged after 300 steps
|
57 |
+
- **Domain Accuracy**: Specialized for Isaac Sim robotics
|
58 |
+
- **Code Quality**: Generated code follows Isaac Sim best practices
|
59 |
+
- **Response Relevance**: High relevance to robotics queries
|
60 |
+
|
61 |
+
### Limitations
|
62 |
+
1. **Domain Specificity**: Limited to Isaac Sim robotics context
|
63 |
+
2. **GGUF Conversion**: Currently has metadata compatibility issues
|
64 |
+
3. **Hardware Requirements**: Requires significant VRAM for full precision
|
65 |
+
4. **Training Data Size**: Limited to 2,000 examples
|
66 |
+
|
67 |
+
### Known Issues
|
68 |
+
- **GGUF Loading Error**: Missing `qwen2.context_length` metadata field
|
69 |
+
- **Workaround**: Use HuggingFace or CTransformers formats
|
70 |
+
- **Status**: Under investigation for future updates
|
71 |
+
|
72 |
+
## Usage
|
73 |
+
|
74 |
+
### Input Format
|
75 |
+
The model expects Isaac Sim-specific queries in the following format:
|
76 |
+
```
|
77 |
+
<|im_start|>user
|
78 |
+
[Your Isaac Sim robotics question here]
|
79 |
+
<|im_end|>
|
80 |
+
<|im_start|>assistant
|
81 |
+
```
|
82 |
+
|
83 |
+
### Example Queries
|
84 |
+
1. **Robot Creation**: "How do I create a differential drive robot in Isaac Sim?"
|
85 |
+
2. **Sensor Setup**: "How to add a depth camera and process depth data?"
|
86 |
+
3. **Physics Configuration**: "What physics parameters should I use for a manipulator?"
|
87 |
+
4. **Environment Design**: "How to create a warehouse environment with obstacles?"
|
88 |
+
5. **Troubleshooting**: "Why is my robot falling through the ground?"
|
89 |
+
|
90 |
+
### Output Characteristics
|
91 |
+
- **Code Generation**: Python scripts ready for Isaac Sim
|
92 |
+
- **Explanation**: Detailed step-by-step instructions
|
93 |
+
- **Best Practices**: Follows Isaac Sim development guidelines
|
94 |
+
- **Error Prevention**: Includes common pitfalls and solutions
|
95 |
+
|
96 |
+
## Model Variants
|
97 |
+
|
98 |
+
### 1. HuggingFace Format (Primary)
|
99 |
+
- **Location**: `models/huggingface/`
|
100 |
+
- **Size**: 5.3GB
|
101 |
+
- **Format**: Standard HuggingFace model files
|
102 |
+
- **Usage**: Direct integration with transformers library
|
103 |
+
- **Advantages**: Full compatibility, easy integration
|
104 |
+
|
105 |
+
### 2. CTransformers Format (Alternative)
|
106 |
+
- **Location**: `models/ctransformers/`
|
107 |
+
- **Size**: 5.2GB
|
108 |
+
- **Format**: Optimized for CTransformers library
|
109 |
+
- **Usage**: Lightweight inference with reduced memory
|
110 |
+
- **Advantages**: Lower memory usage, faster inference
|
111 |
+
|
112 |
+
### 3. GGUF Format (Experimental)
|
113 |
+
- **Location**: `models/gguf/`
|
114 |
+
- **Size**: 616MB (base) + quantization variants
|
115 |
+
- **Format**: llama.cpp compatible
|
116 |
+
- **Usage**: Server deployment and edge inference
|
117 |
+
- **Status**: Metadata issues, conversion scripts provided
|
118 |
+
|
119 |
+
## Ethical Considerations
|
120 |
+
|
121 |
+
### Bias and Fairness
|
122 |
+
- **Training Data**: Focused on technical robotics content
|
123 |
+
- **Domain Limitation**: May not generalize to other robotics platforms
|
124 |
+
- **Cultural Bias**: Minimal, focused on technical accuracy
|
125 |
+
|
126 |
+
### Safety
|
127 |
+
- **Content Filtering**: No additional safety filters applied
|
128 |
+
- **Use Case**: Intended for robotics development only
|
129 |
+
- **Misuse Prevention**: Technical domain limits potential misuse
|
130 |
+
|
131 |
+
### Privacy
|
132 |
+
- **Training Data**: Synthetic data, no personal information
|
133 |
+
- **Inference**: No data collection or logging
|
134 |
+
- **Compliance**: Follows standard AI model privacy practices
|
135 |
+
|
136 |
+
## Technical Specifications
|
137 |
+
|
138 |
+
### Model Architecture
|
139 |
+
- **Base**: Qwen2.5-Coder-7B-Instruct
|
140 |
+
- **Parameters**: 7 billion
|
141 |
+
- **Context Length**: 32,768 tokens
|
142 |
+
- **Vocabulary**: 151,936 tokens
|
143 |
+
- **Embedding Dimension**: 4,096
|
144 |
+
- **Attention Heads**: 32
|
145 |
+
- **Layers**: 32
|
146 |
+
|
147 |
+
### Quantization Support
|
148 |
+
- **FP16**: Full precision (default)
|
149 |
+
- **INT8**: 8-bit quantization support
|
150 |
+
- **INT4**: 4-bit quantization (experimental)
|
151 |
+
- **GGUF**: Conversion scripts provided
|
152 |
+
|
153 |
+
### Integration
|
154 |
+
- **HuggingFace**: Native support
|
155 |
+
- **Isaac Sim**: Direct Python integration
|
156 |
+
- **CTransformers**: Optimized inference
|
157 |
+
- **llama.cpp**: When GGUF issues resolved
|
158 |
+
|
159 |
+
## Deployment
|
160 |
+
|
161 |
+
### Local Development
|
162 |
+
```bash
|
163 |
+
# Clone repository
|
164 |
+
git clone https://github.com/your-username/isaac-sim-robotics-qwen.git
|
165 |
+
cd isaac-sim-robotics-qwen
|
166 |
+
|
167 |
+
# Install dependencies
|
168 |
+
pip install -r requirements.txt
|
169 |
+
|
170 |
+
# Download model
|
171 |
+
huggingface-cli download your-username/isaac-sim-robotics-qwen
|
172 |
+
```
|
173 |
+
|
174 |
+
### Production Deployment
|
175 |
+
- **HuggingFace Hub**: Direct model hosting
|
176 |
+
- **Docker**: Containerized deployment
|
177 |
+
- **API Server**: RESTful inference endpoints
|
178 |
+
- **Edge Deployment**: GGUF format (when fixed)
|
179 |
+
|
180 |
+
## Maintenance
|
181 |
+
|
182 |
+
### Updates
|
183 |
+
- **Training Data**: Expandable dataset for future versions
|
184 |
+
- **Model Architecture**: Base model updates as available
|
185 |
+
- **Bug Fixes**: Regular repository updates
|
186 |
+
- **Community**: Open source maintenance
|
187 |
+
|
188 |
+
### Support
|
189 |
+
- **Documentation**: Comprehensive guides and examples
|
190 |
+
- **Issues**: GitHub issue tracking
|
191 |
+
- **Discussions**: Community support forum
|
192 |
+
- **Examples**: Working code samples
|
193 |
+
|
194 |
+
## Citation
|
195 |
+
|
196 |
+
If you use this model in your research or development, please cite:
|
197 |
+
|
198 |
+
```bibtex
|
199 |
+
@misc{qwen2.5_coder_7b_instruct_omni1.1,
|
200 |
+
title={Qwen2.5-Coder-7B-Instruct-Omni1.1: Isaac Sim Robotics Specialized Model},
|
201 |
+
author={TomBombadyl},
|
202 |
+
year={2025},
|
203 |
+
url={https://huggingface.co/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.1}
|
204 |
+
}
|
205 |
+
```
|
206 |
+
|
207 |
+
## License
|
208 |
+
|
209 |
+
This model is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.
|
210 |
+
|
211 |
+
## Contact
|
212 |
+
|
213 |
+
- **Repository**: [Hugging Face Hub](https://huggingface.co/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.1)
|
214 |
+
- **Issues**: [GitHub Issues](https://github.com/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.1/issues)
|
215 |
+
- **Discussions**: [GitHub Discussions](https://github.com/TomBombadyl/Qwen2.5-Coder-7B-Instruct-Omni1.1/discussions)
|
216 |
+
|
217 |
+
---
|
218 |
+
|
219 |
+
**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.
|