| name | identify-architecture |
| description | Analyze ML model architecture from papers and code. Use when understanding model structure for implementation. |
| mcp_fallback | none |
| category | analysis |
| tier | 2 |
Identify Architecture
Analyze and document machine learning model architectures including layers, connections, and information flow.
When to Use
- Understanding paper model designs
- Planning model implementation
- Comparing architecture variations
- Documenting neural network structure
Quick Reference
# Extract architecture from paper
# Look for: "Figure X: Architecture of [Model]"
# Check for: Table with layer specifications
# Find: Layer descriptions (Conv2D, FC, BatchNorm, etc.)
# Visualize model structure (Mojo)
# var model: SimpleNet = ...
# print(model) # Should show layer information
Workflow
- Locate architecture diagram: Find visual architecture representation in paper
- List layers: Enumerate all layers with type and parameters
- Document connections: Map data flow between layers (skip connections, merges)
- Extract layer parameters: For each layer record size, activation, normalization
- Create implementation plan: Translate to Mojo struct/function definitions
Output Format
Architecture documentation:
- Model name and source
- Layer-by-layer breakdown
- Layer type (Conv2D, Dense, etc.)
- Parameters (kernel size, stride, padding, activation)
- Input/output shapes
- Data flow diagram (text or ASCII)
- Special components (skip connections, attention)
References
- See
extract-hyperparametersskill for model configuration - See CLAUDE.md > Mojo Syntax Standards for implementation patterns
- See
/notes/review/mojo-ml-patterns.mdfor architecture patterns