Files
zeekayandHanzo Dev 02b3f00cef chore: complete brand scrub of upstream mistral.rs lineage to hanzo/hanzoai
Replace all remaining references to the upstream project/author/framework
brands with hanzo / hanzoai across source, comments, docs, examples, and
build scripts.

- Project spellings mistral.rs / mistral-rs / mistralrs -> hanzo (and the
  hanzo-* crate/binary names in Dockerfiles, Makefile, install.ps1, .gitignore).
- Author EricLBuehler / EricBuehler / EricB / Eric Buehler -> hanzoai / Hanzo AI
  (incl. Cargo `authors` and LICENSE copyright).
- URLs: github.com/EricLBuehler/mistral.rs -> github.com/hanzoai/engine;
  ericlbuehler.github.io/mistral.rs -> hanzoai.github.io/engine;
  github.com/huggingface/candle -> github.com/hanzoai/ml;
  docs.rs/candle-* -> docs.rs/hanzo-ml; ghcr.io image -> ghcr.io/hanzoai/engine.
- Env var prefix MISTRALRS_ -> HANZO_ (producers and consumers kept in sync,
  incl. the build-script-injected HANZO_GIT_REVISION).
- ROCm static lib libmistralrsrocm.a / mistralrsrocm -> libhanzorocm.a / hanzorocm.
- localStorage keys mistralrs_settings -> hanzo_settings.

Deliberately preserved: real model architectures (struct Mistral, models/mistral.rs
file, Mixtral/Llama/Gemma/Qwen/Phi), real model-org HF repo-ids (mistralai/,
meta-llama/, google/, Qwen/, microsoft/), the `use hanzo_ml as candle` code
aliases and `candle::` identifiers, and the upstream-author HF data repo-id
suffix in "hanzoai/mistralrs_tests".

Both `cargo check` and `cargo check --features vulkan` pass with 0 errors.

Co-authored-by: Hanzo Dev <[email protected]>
2026-06-01 11:53:12 -07:00

5.8 KiB

Llama 3.2 Vision Model: meta-llama/Llama-3.2-11B-Vision-Instruct

Hanzo Engine supports the Llama 3.2 multimodal model. ISQ quantization is supported to allow running the model with less memory requirements.

UQFF quantizations are also available.

Note: Some examples use the Cephalo Llama 3.2 model, a member of the Cephalo model collection. This model is finetune of Llama 3.2 with enhanced capabilities in scientific images. To use the base Llama 3.2 Vision model, simply use the associated model ID.

Note: When using device mapping or model topology, only the text model and its layers will be managed. This is because it contains most of the model parameters. The text model has 40 layers.

Quick Start

hanzo run -m meta-llama/Llama-3.2-11B-Vision-Instruct --isq 4 --image photo.jpg -i "Describe this image"

Input Formats

The Python and HTTP APIs support sending inputs as:

  • Images: URL, path to a local file, or base64 encoded string (via image_url)

The Rust SDK takes images from the image crate.

UQFF models

UQFF is a quantized file format similar to GGUF based on ISQ. It removes the memory and compute requirements that come with ISQ by providing ready-made quantizations. The key advantage over GGUF is the flexibility to store multiple quantizations in one file.

We provide UQFF files (hanzoai/Llama-3.2-11B-Vision-Instruct-UQFF) for this Llama 3.2 Vision model.

For example:

hanzo run -m meta-llama/Llama-3.2-11B-Vision-Instruct --from-uqff hanzoai/Llama-3.2-11B-Vision-Instruct-UQFF/llama-3.2-11b-vision-q4k.uqff

HTTP API

  1. Start the server
hanzo serve -m meta-llama/Llama-3.2-11B-Vision-Instruct --isq 4 -p 1234
  1. Send a request
from openai import OpenAI

client = OpenAI(api_key="foobar", base_url="http://localhost:1234/v1/")

completion = client.chat.completions.create(
    model="default",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://www.nhmagazine.com/content/uploads/2019/05/mtwashingtonFranconia-2-19-18-108-Edit-Edit.jpg"
                    },
                },
                {
                    "type": "text",
                    "text": "What is this?",
                },
            ],
        },
    ],
    max_tokens=256,
)
print(completion.choices[0].message.content)

Python SDK

You can find this example here.

from hanzo import Runner, Which, ChatCompletionRequest, MultimodalArchitecture

MODEL_ID = "lamm-mit/Cephalo-Llama-3.2-11B-Vision-Instruct-128k"

runner = Runner(
    which=Which.MultimodalPlain(
        model_id=MODEL_ID,
        arch=MultimodalArchitecture.VLlama,
    ),
)

res = runner.send_chat_completion_request(
    ChatCompletionRequest(
        model="default",
        messages=[
            {
                "role": "user",
                "content": [
                    {
                        "type": "image_url",
                        "image_url": {
                            "url": "https://www.nhmagazine.com/content/uploads/2019/05/mtwashingtonFranconia-2-19-18-108-Edit-Edit.jpg"
                        },
                    },
                    {
                        "type": "text",
                        "text": "What is this?",
                    },
                ],
            }
        ],
        max_tokens=256,
    )
)
print(res.choices[0].message.content)
print(res.usage)

Rust SDK

You can find this example here.

use anyhow::Result;
use hanzo::{IsqType, TextMessageRole, MultimodalMessages, MultimodalModelBuilder};

const MODEL_ID: &str = "lamm-mit/Cephalo-Llama-3.2-11B-Vision-Instruct-128k";

#[tokio::main]
async fn main() -> Result<()> {
    let model =
        MultimodalModelBuilder::new(MODEL_ID)
            .with_isq(IsqType::Q4K)
            .with_logging()
            .build()
            .await?;

    let bytes = match reqwest::blocking::get(
        "https://cdn.britannica.com/45/5645-050-B9EC0205/head-treasure-flower-disk-flowers-inflorescence-ray.jpg",
    ) {
        Ok(http_resp) => http_resp.bytes()?.to_vec(),
        Err(e) => anyhow::bail!(e),
    };
    let image = image::load_from_memory(&bytes)?;

    let messages = MultimodalMessages::new().add_image_message(
        TextMessageRole::User,
        "What is depicted here? Please describe the scene in detail.",
        vec![image],
    );

    let response = model.send_chat_request(messages).await?;

    println!("{}", response.choices[0].message.content.as_ref().unwrap());
    dbg!(
        response.usage.avg_prompt_tok_per_sec,
        response.usage.avg_compl_tok_per_sec
    );

    Ok(())
}