# xiaomimimo/mimo architecture

> Architecture diagram of the GitHub repository xiaomimimo/mimo, 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/xiaomimimo/mimo
- Repository: https://github.com/xiaomimimo/mimo

## Overview

MiMo-7B is a reasoning-model project documented from pretraining through supervised fine-tuning and reinforcement learning. Its README describes data preparation, training objectives, rule-based rewards, and rollout infrastructure; the only sampled runtime source registers a MiMo causal-language model implementation with vLLM. Training-stage links below reflect README statements, not inspected training code, so their nodes have no implementation path. The repository does not show the full training or serving workflow.

Main entry points: [`registry/register_mimo_in_vllm.py`](https://github.com/xiaomimimo/mimo/blob/main/registry/register_mimo_in_vllm.py) (MiMo model adapter)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_training["Pretraining"]
  node_pretrain_data["Reasoning data"]
  node_pretrain["Base pretraining"]
  node_mtp["MTP objective"]
  node_base_model["Base model"]
end

subgraph group_posttraining["Posttraining"]
  node_sft["Supervised tuning"]
  node_rl_data["RL problems"]
  node_verifiers["Rule verifiers"]
  node_code_reward["Code rewards"]
  node_resampling["Easy resampling"]
  node_rollout["Rollout engine"]
  node_rl_model["RL model"]
  node_sft_model["SFT model"]
end

subgraph group_serving["Inference"]
  node_mimo_impl["MiMo model adapter"]
  node_vllm["vLLM runtime"]
end

node_researchers(("Model developers"))
node_model_users(("Model users"))

node_researchers -->|"curate data"| node_pretrain_data
node_pretrain_data -->|"trains on"| node_pretrain
node_pretrain -->|"produces"| node_base_model
node_mtp -->|"adds objective"| node_pretrain
node_base_model -->|"starts tuning"| node_sft
node_sft -->|"produces"| node_sft_model
node_base_model -->|"starts RL"| node_rollout
node_sft_model -->|"starts RL"| node_rollout
node_rl_data -->|"verified by"| node_verifiers
node_verifiers -->|"scores answers"| node_rollout
node_code_reward -->|"shapes rewards"| node_rollout
node_resampling -->|"improves sampling"| node_rollout
node_rollout -->|"trains"| node_rl_model
node_mimo_impl -->|"registers model"| node_vllm
node_model_users -.->|"request inference"| node_vllm

click node_mimo_impl "https://github.com/xiaomimimo/mimo/blob/main/registry/register_mimo_in_vllm.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_pretrain_data,node_pretrain,node_mtp,node_base_model,node_model_users toneBlue
class node_sft,node_rl_data,node_verifiers,node_code_reward,node_resampling,node_rollout,node_rl_model,node_sft_model toneAmber
class node_mimo_impl,node_vllm toneMint
class node_researchers toneIndigo
```

## Components

### Pretraining

- **Reasoning data**
- **Base pretraining**
- **MTP objective**
- **Base model**

### Posttraining

- **Supervised tuning**
- **RL problems**
- **Rule verifiers**
- **Code rewards**
- **Easy resampling**
- **Rollout engine**
- **RL model**
- **SFT model**

### Inference

- [MiMo model adapter](https://github.com/xiaomimimo/mimo/blob/main/registry/register_mimo_in_vllm.py): `registry/register_mimo_in_vllm.py`
- **vLLM runtime**

### Other components

- **Model developers**
- **Model users**

## Connections

- Model developers → Reasoning data: curate data (evidence: `README.md`)
- Reasoning data → Base pretraining: trains on (evidence: `README.md`)
- Base pretraining → Base model: produces (evidence: `README.md`)
- MTP objective → Base pretraining: adds objective (evidence: `README.md`)
- Base model → Supervised tuning: starts tuning (evidence: `README.md`)
- Supervised tuning → SFT model: produces (evidence: `README.md`)
- Base model → Rollout engine: starts RL (evidence: `README.md`)
- SFT model → Rollout engine: starts RL (evidence: `README.md`)
- RL problems → Rule verifiers: verified by (evidence: `README.md`)
- Rule verifiers → Rollout engine: scores answers (evidence: `README.md`)
- Code rewards → Rollout engine: shapes rewards (evidence: `README.md`)
- Easy resampling → Rollout engine: improves sampling (evidence: `README.md`)
- Rollout engine → RL model: trains (evidence: `README.md`)
- MiMo model adapter → vLLM runtime: registers model (evidence: `registry/register_mimo_in_vllm.py`)
- Model users → vLLM runtime: request inference
