Instructions to use Cloudflare/clef with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Cloudflare/clef with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="Cloudflare/clef")
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("Cloudflare/clef")
model = AutoModelForMultimodalLM.from_pretrained("Cloudflare/clef", 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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Cloudflare/clef with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Cloudflare/clef"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Cloudflare/clef",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Cloudflare/clef

SGLang

How to use Cloudflare/clef with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Cloudflare/clef" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Cloudflare/clef",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Cloudflare/clef" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Cloudflare/clef",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use Cloudflare/clef with Docker Model Runner:
```
docker model run hf.co/Cloudflare/clef
```

areneau commited on 17 days ago

Commit

7510eee

verified ·

1 Parent(s): a20e258

Decision Index: drop # column, add latency rows

Browse files

Files changed (1) hide show

README.md +47 -45

README.md CHANGED Viewed

@@ -155,51 +155,53 @@ Each question has:
 ### Decision Index
-Per-benchmark results from our internal run of the [Decision Index](https://huggingface.co/spaces/multimodalart/jev-decision-index) 0.2.1 suite. Scores are percentages; ForecastBench is a Brier score, where lower is better. The best score in each row is in bold.
-| # | Benchmark | Clef | Clef-flash | Jev | DiffusionGemma Jev | Kev 9B | Laya |
-|---|---|---|---|---|---|---|---|
-| 1 | BFCL (case exact accuracy) | 98.5 | **98.8** | 95.8 | 96.5 | 94.5 | 38.1 |
-| 2 | ToolRet (nDCG@10) | **69.2** | 66.4 | 65.3 | 61.2 | 64.3 | 12.8 |
-| 3 | API-Bank (accuracy) | 91.9 | **93.1** | 88.2 | 83.7 | 56.3 | 11.5 |
-| 4 | BANKING77 (macro-F1) | **94.2** | 90.9 | 79.7 | 74.3 | 84.8 | 14.3 |
-| 5 | CLINC150+OOS (macro-F1) | **97.4** | 66.8 | 89.3 | 83.5 | 79.0 | 3.2 |
-| 6 | RouterBench (selected quality) | 79.7 | 79.9 | 79.9 | 79.0 | **80.0** | 57.1 |
-| 9 | Home appliance simulator (case exact accuracy) | 83.0 | **97.7** | 52.3 | 42.0 | 25.0 | 0.0 |
-| 10 | SGD/SGD-X (macro-F1) | 43.8 | 34.2 | 43.0 | 40.6 | **64.0** | 42.4 |
-| 11 | ContractNLI (macro-F1) | 81.4 | **84.3** | 71.7 | 76.0 | 57.8 | 29.0 |
-| 12 | ANLI (macro-F1) | 69.8 | 59.1 | **74.8** | 66.4 | 56.3 | 48.7 |
-| 20 | BPoMP (accuracy) | **96.9** | 95.4 | 90.6 | 86.9 | 67.0 | 51.6 |
-| 21 | Humicroedit (accuracy) | 66.7 | **75.1** | 61.9 | 63.0 | 55.8 | 47.2 |
-| 22 | POP909-CL (accuracy) | 15.8 | 1.6 | **18.1** | 2.5 | 10.8 | 5.1 |
-| 23 | cfcolor (accuracy) | **66.0** | 65.8 | 64.7 | 58.2 | 56.3 | 52.3 |
-| 24 | MMLU (accuracy) | 90.3 | **91.8** | 91.7 | 79.3 | 75.3 | 30.7 |
-| 25 | GPQA Diamond (accuracy) | 48.0 | 51.0 | **78.3** | 44.9 | 38.8 | 27.6 |
-| 26 | ARC-Easy (accuracy) | 99.0 | **99.5** | 99.3 | 98.2 | 97.7 | 47.0 |
-| 27 | ARC-Challenge (accuracy) | 97.7 | **98.3** | 97.8 | 94.5 | 93.7 | 28.6 |
-| 28 | WinoGrande (accuracy) | 93.5 | **97.5** | 92.0 | 73.6 | 73.2 | 50.5 |
-| 29 | HellaSwag (accuracy) | 98.2 | **98.6** | 94.5 | 83.3 | 81.9 | 33.1 |
-| 30 | GSM8K (accuracy) | **80.8** | 67.3 | 79.9 | 50.3 | 48.7 | 21.6 |
-| 31 | ChessBench (accuracy) | **24.7** | 23.0 | 17.2 | 14.2 | 11.2 | 7.7 |
-| 32 | MuSR (accuracy) | 83.5 | **86.0** | 66.1 | 61.2 | 57.9 | 43.2 |
-| 33 | SATA-Bench (case exact accuracy) | 33.8 | **36.7** | 26.4 | 27.5 | 26.7 | 0.3 |
-| 36 | BRIGHT (nDCG@10) | 45.9 | 39.3 | **47.5** | 42.9 | 38.5 | 19.9 |
-| 37 | Amazon ESCI (macro-F1) | **57.5** | 57.4 | 55.2 | 53.4 | 49.2 | 24.4 |
-| 38 | ACOS (per-review F1) | **33.3** | 25.9 | 29.5 | 24.5 | 18.3 | 3.5 |
-| 39 | FinEntity (macro-F1) | 96.2 | **97.1** | 87.0 | 89.0 | 88.4 | 61.0 |
-| 41 | VAST (macro-F1) | 59.5 | 49.6 | **64.6** | 55.7 | 55.4 | 40.5 |
-| 42 | NLI4CT (macro-F1) | 82.9 | 78.6 | **84.1** | 78.4 | 74.9 | 47.7 |
-| 43 | CRUXEval (accuracy) | **86.7** | 86.1 | 73.0 | 64.7 | 51.2 | 40.2 |
-| 44 | CLadder (accuracy) | 94.0 | **97.7** | 72.6 | 67.8 | 62.0 | 52.9 |
-| 48 | ForecastBench (Brier, lower is better) | 13.9 | **10.6** | 17.4 | 29.6 | 17.6 | 41.1 |
-| 50 | Habermas Machine (accuracy) | 68.7 | **71.8** | 45.9 | 45.0 | 39.4 | 33.4 |
-| 56 | PhishNChips (accuracy) | 79.6 | 75.0 | 62.5 | **85.4** | 50.7 | 50.1 |
-| 57 | MMLU-Pro (accuracy) | 65.9 | 65.3 | **82.7** | 56.9 | 51.1 | 13.6 |
-| 58 | BBH (accuracy) | 73.7 | 68.9 | **92.9** | 70.7 | 65.2 | 34.1 |
-| 59 | RAGTruth (hallucination F1) | **79.4** | 35.6 | 76.5 | 70.4 | 46.2 | 48.8 |
-| 61 | HoVer (accuracy) | 65.2 | 61.2 | **72.9** | 70.9 | 58.8 | 55.8 |
-| 62 | When2Call MCQ (accuracy) | 72.4 | 65.6 | **81.0** | 75.4 | 49.6 | 11.9 |
-| 64 | New Yorker (accuracy) | 69.5 | 66.1 | **70.1** | 63.6 | 58.1 | 27.1 |
 ### Workflow evals

 ### Decision Index
+Per-benchmark results from our internal run of the [Decision Index](https://huggingface.co/spaces/multimodalart/jev-decision-index) 0.2.1 suite. Scores are percentages; ForecastBench is a Brier score, where lower is better. The last two rows are request latency in milliseconds, where lower is better. The best value in each row is in bold.
+| Benchmark | Clef | Clef-flash | Jev | DiffusionGemma Jev | Kev 9B | Laya |
+|---|---|---|---|---|---|---|
+| BFCL (case exact accuracy) | 98.5 | **98.8** | 95.8 | 96.5 | 94.5 | 38.1 |
+| ToolRet (nDCG@10) | **69.2** | 66.4 | 65.3 | 61.2 | 64.3 | 12.8 |
+| API-Bank (accuracy) | 91.9 | **93.1** | 88.2 | 83.7 | 56.3 | 11.5 |
+| BANKING77 (macro-F1) | **94.2** | 90.9 | 79.7 | 74.3 | 84.8 | 14.3 |
+| CLINC150+OOS (macro-F1) | **97.4** | 66.8 | 89.3 | 83.5 | 79.0 | 3.2 |
+| RouterBench (selected quality) | 79.7 | 79.9 | 79.9 | 79.0 | **80.0** | 57.1 |
+| Home appliance simulator (case exact accuracy) | 83.0 | **97.7** | 52.3 | 42.0 | 25.0 | 0.0 |
+| SGD/SGD-X (macro-F1) | 43.8 | 34.2 | 43.0 | 40.6 | **64.0** | 42.4 |
+| ContractNLI (macro-F1) | 81.4 | **84.3** | 71.7 | 76.0 | 57.8 | 29.0 |
+| ANLI (macro-F1) | 69.8 | 59.1 | **74.8** | 66.4 | 56.3 | 48.7 |
+| BPoMP (accuracy) | **96.9** | 95.4 | 90.6 | 86.9 | 67.0 | 51.6 |
+| Humicroedit (accuracy) | 66.7 | **75.1** | 61.9 | 63.0 | 55.8 | 47.2 |
+| POP909-CL (accuracy) | 15.8 | 1.6 | **18.1** | 2.5 | 10.8 | 5.1 |
+| cfcolor (accuracy) | **66.0** | 65.8 | 64.7 | 58.2 | 56.3 | 52.3 |
+| MMLU (accuracy) | 90.3 | **91.8** | 91.7 | 79.3 | 75.3 | 30.7 |
+| GPQA Diamond (accuracy) | 48.0 | 51.0 | **78.3** | 44.9 | 38.8 | 27.6 |
+| ARC-Easy (accuracy) | 99.0 | **99.5** | 99.3 | 98.2 | 97.7 | 47.0 |
+| ARC-Challenge (accuracy) | 97.7 | **98.3** | 97.8 | 94.5 | 93.7 | 28.6 |
+| WinoGrande (accuracy) | 93.5 | **97.5** | 92.0 | 73.6 | 73.2 | 50.5 |
+| HellaSwag (accuracy) | 98.2 | **98.6** | 94.5 | 83.3 | 81.9 | 33.1 |
+| GSM8K (accuracy) | **80.8** | 67.3 | 79.9 | 50.3 | 48.7 | 21.6 |
+| ChessBench (accuracy) | **24.7** | 23.0 | 17.2 | 14.2 | 11.2 | 7.7 |
+| MuSR (accuracy) | 83.5 | **86.0** | 66.1 | 61.2 | 57.9 | 43.2 |
+| SATA-Bench (case exact accuracy) | 33.8 | **36.7** | 26.4 | 27.5 | 26.7 | 0.3 |
+| BRIGHT (nDCG@10) | 45.9 | 39.3 | **47.5** | 42.9 | 38.5 | 19.9 |
+| Amazon ESCI (macro-F1) | **57.5** | 57.4 | 55.2 | 53.4 | 49.2 | 24.4 |
+| ACOS (per-review F1) | **33.3** | 25.9 | 29.5 | 24.5 | 18.3 | 3.5 |
+| FinEntity (macro-F1) | 96.2 | **97.1** | 87.0 | 89.0 | 88.4 | 61.0 |
+| VAST (macro-F1) | 59.5 | 49.6 | **64.6** | 55.7 | 55.4 | 40.5 |
+| NLI4CT (macro-F1) | 82.9 | 78.6 | **84.1** | 78.4 | 74.9 | 47.7 |
+| CRUXEval (accuracy) | **86.7** | 86.1 | 73.0 | 64.7 | 51.2 | 40.2 |
+| CLadder (accuracy) | 94.0 | **97.7** | 72.6 | 67.8 | 62.0 | 52.9 |
+| ForecastBench (Brier, lower is better) | 13.9 | **10.6** | 17.4 | 29.6 | 17.6 | 41.1 |
+| Habermas Machine (accuracy) | 68.7 | **71.8** | 45.9 | 45.0 | 39.4 | 33.4 |
+| PhishNChips (accuracy) | 79.6 | 75.0 | 62.5 | **85.4** | 50.7 | 50.1 |
+| MMLU-Pro (accuracy) | 65.9 | 65.3 | **82.7** | 56.9 | 51.1 | 13.6 |
+| BBH (accuracy) | 73.7 | 68.9 | **92.9** | 70.7 | 65.2 | 34.1 |
+| RAGTruth (hallucination F1) | **79.4** | 35.6 | 76.5 | 70.4 | 46.2 | 48.8 |
+| HoVer (accuracy) | 65.2 | 61.2 | **72.9** | 70.9 | 58.8 | 55.8 |
+| When2Call MCQ (accuracy) | 72.4 | 65.6 | **81.0** | 75.4 | 49.6 | 11.9 |
+| New Yorker (accuracy) | 69.5 | 66.1 | **70.1** | 63.6 | 58.1 | 27.1 |
+| Median latency (ms) | 209.3 | 38.8 | 524.1 | 84.4 | 51.4 | **5.8** |
+| p95 latency (ms) | 238.6 | **122.4** | 536.0 | 211.2 | 187.9 | 222.5 |
 ### Workflow evals