Instructions to use OpenGVLab/SDLM-3B-D8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OpenGVLab/SDLM-3B-D8 with Transformers:

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

pipe = pipeline("text-generation", model="OpenGVLab/SDLM-3B-D8", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenGVLab/SDLM-3B-D8", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("OpenGVLab/SDLM-3B-D8", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use OpenGVLab/SDLM-3B-D8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OpenGVLab/SDLM-3B-D8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenGVLab/SDLM-3B-D8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OpenGVLab/SDLM-3B-D8

SGLang

How to use OpenGVLab/SDLM-3B-D8 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 "OpenGVLab/SDLM-3B-D8" \
    --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": "OpenGVLab/SDLM-3B-D8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "OpenGVLab/SDLM-3B-D8" \
        --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": "OpenGVLab/SDLM-3B-D8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use OpenGVLab/SDLM-3B-D8 with Docker Model Runner:
```
docker model run hf.co/OpenGVLab/SDLM-3B-D8
```

lll2343

nielsr HF Staff commited on Oct 3, 2025

Commit

04b5afe

verified ·

1 Parent(s): 8082a53

Add project page link to model card (#1)

Browse files

- Add project page link to model card (305dff10dddfd9beae65a6b8a1b6c40c5e9eb558)

Co-authored-by: Niels Rogge <nielsr@users.noreply.huggingface.co>

Files changed (1) hide show

README.md +15 -15

README.md CHANGED Viewed

@@ -1,18 +1,6 @@
 ---
-license: apache-2.0
-license_name: qwen
-license_link: https://huggingface.co/Qwen/Qwen2.5-3B/blob/main/LICENSE
-pipeline_tag: text-generation
-library_name: transformers
 base_model:
 - Qwen/Qwen2.5-3B
-base_model_relation: finetune
-language:
-- en
-tags:
-- sdlm
-- diffusion language model
-- custom_code
 datasets:
 - dyyyyyyyy/ScaleQuest-Math
 - OpenCoder-LLM/opc-sft-stage2
@@ -20,15 +8,27 @@ datasets:
 - HuggingFaceTB/smoltalk2
 - LipengCS/Table-GPT
 - allenai/SciRIFF
 ---
 # SDLM-3B-D8
-[\[📂 GitHub\]](https://github.com/OpenGVLab/SDLM)  [\[📜 Tech Report\]](https://arxiv.org/abs/2509.24007)  [\[🤗 HuggingFace\]](https://huggingface.co/collections/OpenGVLab/sdlm-68ac82709d7c343ad36aa552)
 ## Introduction
-We propose a <b>S</b>equential <b>D</b>iffusion <b>L</b>anguage <b>M</b>odel (<b>SDLM</b>), to cheaply stimulate the parallel prediction capabilities of diffusion models. Specifically, SDLM reduces distribution shift by limiting the prediction range to a fixed block length and enforces decoding order through the longest prefix decoding method, thereby significantly improving prediction efficiency while ensuring generation quality. Our method can be viewed as a further generalization of the autoregressive (AR) paradigm. Therefore, it is possible to use pre-trained AR weights and quickly migrate to the diffusion framework with only minimal instruction fine-tuning.
 ![image/png](https://huggingface.co/OpenGVLab/SDLM-3B-D8/resolve/main/assets/three_framework.png)
@@ -151,4 +151,4 @@ If you find this project useful in your research, please consider citing:
   journal={arXiv preprint arXiv:2509.24007},
   year={2025}
 }
-```

 ---
 base_model:
 - Qwen/Qwen2.5-3B
 datasets:
 - dyyyyyyyy/ScaleQuest-Math
 - OpenCoder-LLM/opc-sft-stage2
 - HuggingFaceTB/smoltalk2
 - LipengCS/Table-GPT
 - allenai/SciRIFF
+language:
+- en
+library_name: transformers
+license: apache-2.0
+license_name: qwen
+license_link: https://huggingface.co/Qwen/Qwen2.5-3B/blob/main/LICENSE
+pipeline_tag: text-generation
+tags:
+- sdlm
+- diffusion language model
+- custom_code
+base_model_relation: finetune
 ---
 # SDLM-3B-D8
+[\[📂 GitHub\]](https://github.com/OpenGVLab/SDLM)  [\[📜 Tech Report\]](https://arxiv.org/abs/2509.24007)  [\\[🚀 Project Page\\]](https://internvl.github.io/blog/2025-09-29-SDLM/)  [\[🤗 HuggingFace\]](https://huggingface.co/collections/OpenGVLab/sdlm-68ac82709d7c343ad36aa552)
 ## Introduction
+We propose a **S**equential **D**iffusion **L**anguage **M**odel (**SDLM**), to cheaply stimulate the parallel prediction capabilities of diffusion models. Specifically, SDLM reduces distribution shift by limiting the prediction range to a fixed block length and enforces decoding order through the longest prefix decoding method, thereby significantly improving prediction efficiency while ensuring generation quality. Our method can be viewed as a further generalization of the autoregressive (AR) paradigm. Therefore, it is possible to use pre-trained AR weights and quickly migrate to the diffusion framework with only minimal instruction fine-tuning.
 ![image/png](https://huggingface.co/OpenGVLab/SDLM-3B-D8/resolve/main/assets/three_framework.png)
   journal={arXiv preprint arXiv:2509.24007},
   year={2025}
 }
+```