How to use from the
Use from the
Transformers library
# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="Level5Tech/clef-NVFP4")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)
# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("Level5Tech/clef-NVFP4")
model = AutoModelForMultimodalLM.from_pretrained("Level5Tech/clef-NVFP4", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))
Quick Links

clef-NVFP4

NVFP4 W4A4, built with llm-compressor, for vLLM. A quantized build of Cloudflare's Clef (27B), made for "The Noise Floor of a Quantization Comparison" by Aleks Blagojevich and Trevor Sunderland (Level5). This isn't an official Cloudflare release.

These builds keep the joint schema head, the vision tower, the norms and the Gated DeltaNet gate projections in BF16. The recipes are in the repository.

Paper: https://arxiv.org/abs/2611.15671

Measurements, recipes and per-record outputs: https://github.com/Level5Tech/clef-quant-bench

The paper and all six builds, in one collection: https://huggingface.co/collections/Level5Tech/the-noise-floor-of-a-quantization-comparison-6afd62e8288cc5e14becefaf

Archived release (all versions): https://doi.org/10.5281/zenodo.24184497

The base model is under Apache-2.0. Its LICENSE is included here, and NOTICE says these weights were modified by quantization.

Citation

@misc{blagojevich2026noisefloorquantizationcomparison,
  title={The Noise Floor of a Quantization Comparison: Engine, restart, batch and kernel drift, measured against 8-bit and 4-bit effects in two decision models},
  author={Aleks Blagojevich and Trevor Sunderland},
  year={2026},
  eprint={2611.15671},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2611.15671},
}
Downloads last month
74
Safetensors
Model size
27B params
Tensor type
BF16
·
U8
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Level5Tech/clef-NVFP4

Base model

Qwen/Qwen3.8-27B
Finetuned
Cloudflare/clef
Quantized
(46)
this model

Collection including Level5Tech/clef-NVFP4

Paper for Level5Tech/clef-NVFP4