Instructions to use maldv/Olmo-3.1-32B-ThinkInstruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use maldv/Olmo-3.1-32B-ThinkInstruct with Transformers:

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

pipe = pipeline("text-generation", model="maldv/Olmo-3.1-32B-ThinkInstruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("maldv/Olmo-3.1-32B-ThinkInstruct")
model = AutoModelForCausalLM.from_pretrained("maldv/Olmo-3.1-32B-ThinkInstruct")
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 maldv/Olmo-3.1-32B-ThinkInstruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "maldv/Olmo-3.1-32B-ThinkInstruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "maldv/Olmo-3.1-32B-ThinkInstruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/maldv/Olmo-3.1-32B-ThinkInstruct

SGLang

How to use maldv/Olmo-3.1-32B-ThinkInstruct 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 "maldv/Olmo-3.1-32B-ThinkInstruct" \
    --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": "maldv/Olmo-3.1-32B-ThinkInstruct",
		"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 "maldv/Olmo-3.1-32B-ThinkInstruct" \
        --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": "maldv/Olmo-3.1-32B-ThinkInstruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use maldv/Olmo-3.1-32B-ThinkInstruct with Docker Model Runner:
```
docker model run hf.co/maldv/Olmo-3.1-32B-ThinkInstruct
```

Model Details

Model Card for Olmo-3.1-32B-ThinkInstruct

AllenAI dropped a pretty good Think and Instruct model based on two very different branches off of Olmo 3. I merged them together using my normal bespoke fourier interpolation process, but kept the last two layers and the lm head from Think otherwise it would forget to close the think tag.

It's completely coherent. It tends to think a lot less than the standard Think model.

License

This model is licensed under Apache 2.0.

Downloads last month: 5

Safetensors

Model size

32B params

Tensor type

BF16

Model tree for maldv/Olmo-3.1-32B-ThinkInstruct

Base model

allenai/Olmo-3-1125-32B

Finetuned

(5)

this model