# hzlbbfrog/crackmamba architecture

> Architecture diagram of the GitHub repository hzlbbfrog/crackmamba, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-01). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/hzlbbfrog/crackmamba
- Repository: https://github.com/hzlbbfrog/crackmamba

## Overview

The checked-in runtime is the vendored Mamba 1.2.0 language-model library; the README describes CrackMamba crack segmentation, but this tree contains no segmentation model or image-to-mask workflow. The represented runtime configures a Mamba language model, runs its selective state-space operations, and supports token generation and Hugging Face checkpoint loading. The README’s installation and Jetson tutorials are not runtime components. Connections are limited to relationships visible in excerpts or the source index.

Main entry points: [`mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py) (LM model), [`mamba-1.2.0/mamba_ssm/modules/mamba_simple.py`](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/modules/mamba_simple.py) (Mamba block), [`mamba-1.2.0/mamba_ssm/ops/selective_scan_interface.py`](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/selective_scan_interface.py) (Scan interface), [`mamba-1.2.0/mamba_ssm/utils/generation.py`](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/utils/generation.py) (Token generation)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_model["Model assembly"]
  node_config["Model config<br/>[config_mamba.py]"]
  node_lm["LM model"]
  node_mixer["Mamba block<br/>[mamba_simple.py]"]
end

subgraph group_ops["State-space operations"]
  node_norm["Triton norms<br/>[layernorm.py]"]
  node_scan_api["Scan interface"]
  node_state_update["State update"]
end

subgraph group_inference["Inference support"]
  node_generation["Token generation<br/>[generation.py]"]
  node_hf["HF checkpoint loading<br/>[hf.py]"]
  node_eval["LM evaluation adapter<br/>[lm_harness_eval.py]"]
end

subgraph group_native["CUDA extension"]
  node_scan_cpp["Scan bindings<br/>[selective_scan.cpp]"]
  node_scan_header["Scan parameters<br/>[selective_scan.h]"]
end

node_caller(("Model user"))
node_hub[("Hugging Face Hub")]

node_caller -.->|"uses model"| node_lm
node_lm -->|"configures from"| node_config
node_lm -->|"builds blocks"| node_mixer
node_lm -->|"uses norms"| node_norm
node_mixer -->|"runs scan"| node_scan_api
node_mixer -->|"updates state"| node_state_update
node_mixer -->|"uses norms"| node_norm
node_scan_api -->|"calls extension"| node_scan_cpp
node_scan_cpp -->|"uses parameter types"| node_scan_header
node_lm -->|"inherits generation"| node_generation
node_generation -->|"runs model"| node_lm
node_lm -->|"loads checkpoints"| node_hf
node_hf -->|"fetches config and weights"| node_hub
node_eval -->|"loads pretrained model"| node_lm

click node_config "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/models/config_mamba.py"
click node_lm "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py"
click node_mixer "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/modules/mamba_simple.py"
click node_norm "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/triton/layernorm.py"
click node_scan_api "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/selective_scan_interface.py"
click node_scan_cpp "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/csrc/selective_scan/selective_scan.cpp"
click node_scan_header "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/csrc/selective_scan/selective_scan.h"
click node_state_update "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/triton/selective_state_update.py"
click node_generation "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/utils/generation.py"
click node_hf "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/utils/hf.py"
click node_eval "https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/evals/lm_harness_eval.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_config,node_lm,node_mixer,node_caller toneBlue
class node_norm,node_scan_api,node_state_update,node_hub toneAmber
class node_generation,node_hf,node_eval toneMint
class node_scan_cpp,node_scan_header toneRose
```

## Components

### Model assembly

- [Model config](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/models/config_mamba.py): `mamba-1.2.0/mamba_ssm/models/config_mamba.py`
- [LM model](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py): `mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`
- [Mamba block](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/modules/mamba_simple.py): `mamba-1.2.0/mamba_ssm/modules/mamba_simple.py`

### State-space operations

- [Triton norms](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/triton/layernorm.py): `mamba-1.2.0/mamba_ssm/ops/triton/layernorm.py`
- [Scan interface](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/selective_scan_interface.py): `mamba-1.2.0/mamba_ssm/ops/selective_scan_interface.py`
- [State update](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/ops/triton/selective_state_update.py): `mamba-1.2.0/mamba_ssm/ops/triton/selective_state_update.py`

### Inference support

- [Token generation](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/utils/generation.py): `mamba-1.2.0/mamba_ssm/utils/generation.py`
- [HF checkpoint loading](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/mamba_ssm/utils/hf.py): `mamba-1.2.0/mamba_ssm/utils/hf.py`
- [LM evaluation adapter](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/evals/lm_harness_eval.py): `mamba-1.2.0/evals/lm_harness_eval.py`

### CUDA extension

- [Scan bindings](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/csrc/selective_scan/selective_scan.cpp): `mamba-1.2.0/csrc/selective_scan/selective_scan.cpp`
- [Scan parameters](https://github.com/hzlbbfrog/crackmamba/blob/main/mamba-1.2.0/csrc/selective_scan/selective_scan.h): `mamba-1.2.0/csrc/selective_scan/selective_scan.h`

### Other components

- **Model user**
- **Hugging Face Hub**

## Connections

- Model user → LM model: uses model
- LM model → Model config: configures from (evidence: `mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`)
- LM model → Mamba block: builds blocks (evidence: `mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`)
- LM model → Triton norms: uses norms (evidence: `mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`)
- Mamba block → Scan interface: runs scan (evidence: `mamba-1.2.0/mamba_ssm/modules/mamba_simple.py`)
- Mamba block → State update: updates state (evidence: `mamba-1.2.0/mamba_ssm/modules/mamba_simple.py`)
- Mamba block → Triton norms: uses norms (evidence: `mamba-1.2.0/mamba_ssm/modules/mamba_simple.py`)
- Scan interface → Scan bindings: calls extension (evidence: `mamba-1.2.0/mamba_ssm/ops/selective_scan_interface.py`)
- Scan bindings → Scan parameters: uses parameter types (evidence: `mamba-1.2.0/csrc/selective_scan/selective_scan.cpp`)
- LM model → Token generation: inherits generation (evidence: `mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`)
- Token generation → LM model: runs model (evidence: `mamba-1.2.0/mamba_ssm/utils/generation.py`)
- LM model → HF checkpoint loading: loads checkpoints (evidence: `mamba-1.2.0/mamba_ssm/models/mixer_seq_simple.py`)
- HF checkpoint loading → Hugging Face Hub: fetches config and weights (evidence: `mamba-1.2.0/mamba_ssm/utils/hf.py`)
- LM evaluation adapter → LM model: loads pretrained model (evidence: `mamba-1.2.0/evals/lm_harness_eval.py`)
