# mode-io/vllm-jev architecture

> Architecture diagram of the GitHub repository mode-io/vllm-jev, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-06). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/mode-io/vllm-jev
- Repository: https://github.com/mode-io/vllm-jev

## Overview

vLLM Jev serves structured Choice, Noul, and Score decisions from Jev-style checkpoints. The main workflow is a user starting a supported model, then submitting a decision request to an HTTP endpoint, which dispatches to the checkpoint-specific inference service and returns candidate scores and a selected label. Linux uses native vLLM; Apple Silicon has separate serving paths. The graph also preserves documented multimodal and model-family capabilities. Some subsystem wiring is inferred only from the SOURCE INDEX references, so uncertain links are omitted.

Main entry points: [`vllm_jev/endpoint.py`](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/endpoint.py) (HTTP endpoint), [`vllm_jev/checkpoint.py`](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/checkpoint.py) (Checkpoint verification), [`vllm_jev/valen.py`](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/valen.py) (Valen inference), [`vllm_jev/media.py`](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/media.py) (Media loading)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_serving["Serving"]
  node_cli["Serve CLI<br/>[cli.py]"]
  node_endpoint["HTTP endpoint<br/>[endpoint.py]"]
  node_mac["Mac serving<br/>[mac.py]"]
  node_mac_laya["Mac Laya<br/>[mac_laya.py]"]
  node_mac_valen["Mac Valen<br/>[mac_valen.py]"]
end

subgraph group_decisions["Decision models"]
  node_protocols["Decision protocols"]
  node_decision["Typed decisions<br/>[decision.py]"]
  node_laya["Laya inference<br/>[laya.py]"]
  node_laya_graph["Laya readout<br/>[laya_graph.py]"]
end

subgraph group_multimodal["Multimodal models"]
  node_media["Media loading<br/>[media.py]"]
  node_video["Video loading<br/>[video.py]"]
  node_valen["Valen inference<br/>[valen.py]"]
  node_clef["Clef inference<br/>[clef.py]"]
  node_rsijev["RSI-Jev inference<br/>[rsijev.py]"]
  node_vjev["vjev inference<br/>[vjev.py]"]
end

subgraph group_checkpointing["Checkpoint preparation"]
  node_checkpoint["Checkpoint verification<br/>[checkpoint.py]"]
  node_decision_export["Decision export<br/>[decision_export.py]"]
  node_laya_export["Laya export<br/>[laya_export.py]"]
  node_clef_export["Clef export<br/>[clef_export.py]"]
  node_rsijev_export["RSI-Jev export<br/>[rsijev_export.py]"]
  node_valen_export["Valen export<br/>[valen_export.py]"]
  node_vjev_export["vjev export<br/>[vjev_export.py]"]
end

node_user(("Model user"))

node_user -->|"starts model"| node_cli
node_user -->|"submits decisions"| node_endpoint
node_endpoint -->|"dispatches"| node_decision
node_endpoint -->|"dispatches"| node_laya
node_endpoint -->|"dispatches"| node_valen
node_endpoint -->|"dispatches"| node_clef
node_endpoint -->|"dispatches"| node_rsijev
node_endpoint -->|"dispatches"| node_vjev
node_endpoint -->|"uses"| node_protocols
node_laya -->|"uses readout"| node_laya_graph
node_mac -->|"attaches router"| node_endpoint
node_mac_laya -->|"uses serving"| node_mac
node_mac_valen -->|"uses serving"| node_mac
node_mac_laya -->|"uses inference"| node_laya
node_mac_valen -->|"uses inference"| node_valen
node_valen -->|"loads images"| node_media
node_valen -->|"loads video"| node_video
node_clef -->|"loads media"| node_media
node_clef -->|"loads video"| node_video
node_rsijev -->|"loads images"| node_media
node_vjev -->|"loads images"| node_media
node_decision_export -->|"verifies checkpoint"| node_checkpoint
node_laya_export -->|"uses verification"| node_checkpoint
node_clef_export -->|"uses verification"| node_checkpoint
node_rsijev_export -->|"uses verification"| node_checkpoint
node_valen_export -->|"uses verification"| node_checkpoint
node_vjev_export -->|"uses verification"| node_checkpoint

click node_cli "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/cli.py"
click node_endpoint "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/endpoint.py"
click node_mac "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/mac.py"
click node_mac_laya "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/mac_laya.py"
click node_mac_valen "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/mac_valen.py"
click node_protocols "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/decision_protocols.py"
click node_decision "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/decision.py"
click node_laya "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/laya.py"
click node_laya_graph "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/laya_graph.py"
click node_media "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/media.py"
click node_video "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/video.py"
click node_valen "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/valen.py"
click node_clef "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/clef.py"
click node_rsijev "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/rsijev.py"
click node_vjev "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/vjev.py"
click node_checkpoint "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/checkpoint.py"
click node_decision_export "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/decision_export.py"
click node_laya_export "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/laya_export.py"
click node_clef_export "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/clef_export.py"
click node_rsijev_export "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/rsijev_export.py"
click node_valen_export "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/valen_export.py"
click node_vjev_export "https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/vjev_export.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_cli,node_endpoint,node_mac,node_mac_laya,node_mac_valen,node_user toneBlue
class node_protocols,node_decision,node_laya,node_laya_graph toneAmber
class node_media,node_video,node_valen,node_clef,node_rsijev,node_vjev toneMint
class node_checkpoint,node_decision_export,node_laya_export,node_clef_export,node_rsijev_export,node_valen_export,node_vjev_export toneRose
```

## Components

### Serving

- [Serve CLI](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/cli.py): `vllm_jev/cli.py`
- [HTTP endpoint](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/endpoint.py): `vllm_jev/endpoint.py`
- [Mac serving](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/mac.py): `vllm_jev/mac.py`
- [Mac Laya](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/mac_laya.py): `vllm_jev/mac_laya.py`
- [Mac Valen](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/mac_valen.py): `vllm_jev/mac_valen.py`

### Decision models

- [Decision protocols](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/decision_protocols.py): `vllm_jev/decision_protocols.py`
- [Typed decisions](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/decision.py): `vllm_jev/decision.py`
- [Laya inference](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/laya.py): `vllm_jev/laya.py`
- [Laya readout](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/laya_graph.py): `vllm_jev/laya_graph.py`

### Multimodal models

- [Media loading](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/media.py): `vllm_jev/media.py`
- [Video loading](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/video.py): `vllm_jev/video.py`
- [Valen inference](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/valen.py): `vllm_jev/valen.py`
- [Clef inference](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/clef.py): `vllm_jev/clef.py`
- [RSI-Jev inference](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/rsijev.py): `vllm_jev/rsijev.py`
- [vjev inference](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/vjev.py): `vllm_jev/vjev.py`

### Checkpoint preparation

- [Checkpoint verification](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/checkpoint.py): `vllm_jev/checkpoint.py`
- [Decision export](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/decision_export.py): `vllm_jev/decision_export.py`
- [Laya export](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/laya_export.py): `vllm_jev/laya_export.py`
- [Clef export](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/clef_export.py): `vllm_jev/clef_export.py`
- [RSI-Jev export](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/rsijev_export.py): `vllm_jev/rsijev_export.py`
- [Valen export](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/valen_export.py): `vllm_jev/valen_export.py`
- [vjev export](https://github.com/mode-io/vllm-jev/blob/main/vllm_jev/vjev_export.py): `vllm_jev/vjev_export.py`

### Other components

- **Model user**

## Connections

- Model user → Serve CLI: starts model (evidence: `README.md`)
- Model user → HTTP endpoint: submits decisions (evidence: `README.md`)
- HTTP endpoint → Typed decisions: dispatches (evidence: `vllm_jev/endpoint.py`)
- HTTP endpoint → Laya inference: dispatches (evidence: `vllm_jev/endpoint.py`)
- HTTP endpoint → Valen inference: dispatches (evidence: `vllm_jev/endpoint.py`)
- HTTP endpoint → Clef inference: dispatches (evidence: `vllm_jev/endpoint.py`)
- HTTP endpoint → RSI-Jev inference: dispatches (evidence: `vllm_jev/endpoint.py`)
- HTTP endpoint → vjev inference: dispatches (evidence: `vllm_jev/endpoint.py`)
- HTTP endpoint → Decision protocols: uses (evidence: `vllm_jev/decision.py`)
- Laya inference → Laya readout: uses readout (evidence: `vllm_jev/laya.py`)
- Mac serving → HTTP endpoint: attaches router (evidence: `vllm_jev/mac.py`)
- Mac Laya → Mac serving: uses serving (evidence: `vllm_jev/mac_laya.py`)
- Mac Valen → Mac serving: uses serving (evidence: `vllm_jev/mac_valen.py`)
- Mac Laya → Laya inference: uses inference (evidence: `vllm_jev/mac_laya.py`)
- Mac Valen → Valen inference: uses inference (evidence: `vllm_jev/mac_valen.py`)
- Valen inference → Media loading: loads images (evidence: `vllm_jev/valen.py`)
- Valen inference → Video loading: loads video (evidence: `vllm_jev/valen.py`)
- Clef inference → Media loading: loads media (evidence: `vllm_jev/clef.py`)
- Clef inference → Video loading: loads video (evidence: `vllm_jev/clef.py`)
- RSI-Jev inference → Media loading: loads images (evidence: `vllm_jev/rsijev.py`)
- vjev inference → Media loading: loads images (evidence: `vllm_jev/vjev.py`)
- Decision export → Checkpoint verification: verifies checkpoint (evidence: `vllm_jev/decision_export.py`)
- Laya export → Checkpoint verification: uses verification (evidence: `vllm_jev/laya_export.py`)
- Clef export → Checkpoint verification: uses verification (evidence: `vllm_jev/clef_export.py`)
- RSI-Jev export → Checkpoint verification: uses verification (evidence: `vllm_jev/rsijev_export.py`)
- Valen export → Checkpoint verification: uses verification (evidence: `vllm_jev/valen_export.py`)
- vjev export → Checkpoint verification: uses verification (evidence: `vllm_jev/vjev_export.py`)
