Instructions to use ulamai/Ulam-1-Small with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ulamai/Ulam-1-Small with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="ulamai/Ulam-1-Small") 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("ulamai/Ulam-1-Small") model = AutoModelForCausalLM.from_pretrained("ulamai/Ulam-1-Small", 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 ulamai/Ulam-1-Small with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "ulamai/Ulam-1-Small" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "ulamai/Ulam-1-Small", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/ulamai/Ulam-1-Small
- SGLang
How to use ulamai/Ulam-1-Small 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 "ulamai/Ulam-1-Small" \ --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": "ulamai/Ulam-1-Small", "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 "ulamai/Ulam-1-Small" \ --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": "ulamai/Ulam-1-Small", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use ulamai/Ulam-1-Small with Docker Model Runner:
docker model run hf.co/ulamai/Ulam-1-Small
Download RESPONSIBLE_USE.md from ulamai/Ulam-1-Small: direct link, hf CLI and curl.
- Browser
- Download file 1.45 kB
-
https://huggingface.co/ulamai/Ulam-1-Small/resolve/main/RESPONSIBLE_USE.md
- Command line
-
hf download hf://ulamai/Ulam-1-Small/RESPONSIBLE_USE.md
-
curl -L -o RESPONSIBLE_USE.md https://huggingface.co/ulamai/Ulam-1-Small/resolve/main/RESPONSIBLE_USE.md
Responsible use
Ulam-1-Small is a research assistant for mathematical reasoning, not a formal proof checker or autonomous theorem-certification system.
Required human review
Independently check any generated theorem, proof, counterexample, literature claim, or numerical conclusion before relying on it. Claims labeled solved, refuted, likely false, or counterexample found deserve especially strict review. Prefer machine-checkable certificates or formalization where possible.
Output handling
Raw generations may include <think>...</think> spans, partial terminal
answers, or long self-corrections. Public applications should choose and test a
clear policy for retaining, separately routing, or suppressing reasoning spans.
Do not represent hidden or discarded reasoning as independently verified.
Prohibited or unsupported use
- high-stakes medical, legal, financial, or safety decisions;
- unsupervised public theorem announcements;
- autonomous tool or API use without dedicated validation;
- extracting or republishing private benchmark content;
- presenting outputs as citations to sources the model did not inspect;
Evaluation boundaries
Published benchmark figures are checkpoint-selection evidence under their recorded protocols. They are not guarantees for a different prompt template, serving cap, runtime, seed, quantization, or downstream fine-tune. Rerun admission tests after changing any of those conditions.