Instructions to use alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B with Transformers:

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

pipe = pipeline("text-generation", model="alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B")
model = AutoModelForCausalLM.from_pretrained("alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B", device_map="auto")
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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B

SGLang

How to use alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B 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 "alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B" \
    --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": "alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B",
		"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 "alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B" \
        --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": "alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B with Docker Model Runner:
```
docker model run hf.co/alamios/DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

DeepSeek-R1-DRAFT-Qwen2.5-Coder-0.5B

Updated to v1

This model is trained on CODE outputs of deepseek-ai/DeepSeek-R1-Distill-Qwen-32B and is meant to be used only as draft model for speculative decoding.

It's specifically intended for users of 3090/4090, allowing you to run the DeepSeek-R1-Distill-Qwen-32B-Q4_K_M GGUF version with 16k context and speeding up generation without sacrificing more context length or model quality.

Data info

The data consists of code tasks collected from various datasets. It has been trained for 2 epochs on 2.5k unique examples, for a total of 7.6 million tokens per epoch.

Since data generation was done using spare GPU time, I may publish a further trained version later.