---
license: apache-2.0
library_name: transformers
pipeline_tag: image-text-to-text
base_model: Cloudflare/clef-flash
base_model_relation: quantized
tags:
- clef
- cloudflare
- systemone
- qwen3.5
- post-train
- image-text-to-typed-output
- multimodal
- structured-output
- classification
- custom-code
- bitsandbytes
- 4-bit
- nf4
---

# Clef-Flash (4-bit NF4 Quantized)

This repository contains the **4-bit NF4 quantized** version of Cloudflare's **Clef-Flash** multimodal decision model.

- **Base Model:** [Cloudflare/clef-flash](https://huggingface.co/Cloudflare/clef-flash)
- **Quantization:** 4-bit NormalFloat (NF4) with double quantization via `bitsandbytes`
- **Compute Dtype:** `bfloat16`
- **Backbone:** Qwen3.5-9B
- **Joint Schema Head:** Unquantized BF16 precision for accurate scoring and routing
- **Format:** Safetensors

## Quickstart / Usage

```python
import sys
import torch
from huggingface_hub import snapshot_download

path = snapshot_download("meossistant/clef-flash-4bit")
sys.path.insert(0, path)
from joint_schema_model import load_release_model, systemone

model, processor = load_release_model(path, device="cuda")

response = systemone(model, processor, {
    "model": "clef-flash",
    "state": "Our checkout started returning errors and orders are blocked.",
    "questions": {
        "department": {
            "type": "choice",
            "instructions": "Which team should handle the message?",
            "criteria": {"billing": "Payments or invoices", "technical": "Bugs or outages"},
        },
        "urgency": {"type": "score", "criteria": ["Can wait", "This week", "Today"]},
        "outage": {"type": "noul", "instructions": "Is a service down?"},
    },
})
print(response["answers"])
```

## Overview

Clef-Flash is a 9B multimodal model that turns a state and a schema of typed questions into decisions in a single forward pass.
This 4-bit quantized version reduces the VRAM requirement to ~6 GB, making it easily runnable on consumer GPUs.