README.md

---
license: other
license_name: mit-and-llama3.2-community
license_link: https://huggingface.co/amd/Nitro-E
library_name: coreml
pipeline_tag: text-to-image
base_model: amd/Nitro-E
base_model_relation: quantized
tags:
  - coreml
  - core-ml
  - ios
  - macos
  - apple
  - on-device
  - text-to-image
  - diffusion
  - mmdit
  - few-step
---

# Nitro-E (4-Step) — Core ML

*AMD, 2025*

AMD's 304M E-MMDiT text-to-image model (Oct 2025). 4-step distilled variant, 512×512. Llama 3.2 1B text encoder + E-MMDiT denoiser + DC-AE VAE decoder. ~1.04 GB bundled after INT4/INT8 palettization. ~2–3 s / image on iPhone 15+.

<p><img src="https://huggingface.co/mlboydaisuke/Nitro-E-CoreML/resolve/main/media/595b2fd081.png" alt="Nitro-E (4-Step) demo"></p>

Core ML conversion of [amd/Nitro-E](https://huggingface.co/amd/Nitro-E) for on-device inference on iPhone, iPad and Mac. Converted with `coremltools`; the packages are stateless, so all sequencing and buffering lives in your Swift code.

| | |
|---|---|
| Task | text to image |
| Upstream | [amd/Nitro-E](https://huggingface.co/amd/Nitro-E) |
| Packages | 3 |
| Download size | 987 MB |
| Minimum iOS | 18.0 |
| Peak RAM | ~2500 MB |

## Files

| File | Size | Compute units | SHA-256 |
|---|---:|---|---|
| `NitroE_TextEncoder.mlpackage.zip` | 545 MB | `cpuAndNeuralEngine` | `9b366b29d790ab98…` |
| `NitroE_EMMDiT.mlpackage.zip` | 283 MB | `cpuAndNeuralEngine` | `93a7ed971c5c419d…` |
| `NitroE_VAEDecoder.mlpackage.zip` | 153 MB | `cpuAndNeuralEngine` | `4837023736d82b49…` |
| `Llama3Vocab.json` | 2 MB | `-` | `f8f40517934d6f5d…` |
| `Llama3Merges.txt` | 3 MB | `-` | `0cd100e0ab7dbd83…` |
| **Total** | **987 MB** | | |

`compute_units` is not a suggestion -- it is the configuration the conversion was verified against. Moving a package to a different compute unit can silently change the numerics (FP16 attention overflow) or crash on the GPU.

## Download

```bash
hf download mlboydaisuke/coreml-zoo --include "nitroe/*" --local-dir ./nitroe
unzip './nitroe/nitroe/*.zip' -d ./nitroe
```

## Use in Swift

```swift
import CoreML

let config = MLModelConfiguration()
config.computeUnits = .cpuAndNeuralEngine   // as converted — see the table above

// Unzip the .mlpackage, drop it into your Xcode target and Xcode compiles it
// at build time:
let model = try NitroE_TextEncoder(configuration: config)

// ...or compile a downloaded .mlpackage at runtime:
let compiled = try await MLModel.compileModel(at: mlpackageURL)
let model = try MLModel(contentsOf: compiled, configuration: config)
```

> This model is split into 3 Core ML packages that are driven in sequence from Swift. Load them one at a time, copy the outputs out of the `MLMultiArray` buffers and release each model before loading the next — two large Core ML models resident at once will OOM on an iPhone.

## Demo

- **Sample app** — [`sample_apps/NitroEDemo`](https://github.com/john-rocky/CoreML-Models/tree/master/sample_apps/NitroEDemo), a standalone SwiftUI project.
- **Models Zoo** — this model is downloadable and runnable inside the [Models Zoo app](https://apps.apple.com/app/id6762083207) on the App Store, no build required.

## Conversion

- Script: [`convert_nitro_e_emmdit.py`](https://github.com/john-rocky/CoreML-Models/blob/master/conversion_scripts/convert_nitro_e_emmdit.py)
- Pitfalls hit during conversion (FP16 overflow, ANE buffer limits, stride handling): [`docs/coreml_conversion_notes.md`](https://github.com/john-rocky/CoreML-Models/blob/master/docs/coreml_conversion_notes.md)
- Model index: [CoreML-Models](https://github.com/john-rocky/CoreML-Models)

## License

The conversion inherits the upstream license: **MIT (Nitro-E) + Llama 3.2 Community License (text encoder)**.
See [https://huggingface.co/amd/Nitro-E](https://huggingface.co/amd/Nitro-E).

> Nitro-E itself is MIT; the bundled text encoder is Llama 3.2 and carries the Llama 3.2 Community License.

## Credits

- Upstream authors: [amd/Nitro-E](https://huggingface.co/amd/Nitro-E), 2025
- Core ML conversion: john-rocky (Daisuke Majima)