Instructions to use DigitalLearningGmbH/educa-ai-nemo-think with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use DigitalLearningGmbH/educa-ai-nemo-think with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="DigitalLearningGmbH/educa-ai-nemo-think") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("DigitalLearningGmbH/educa-ai-nemo-think") model = AutoModelForCausalLM.from_pretrained("DigitalLearningGmbH/educa-ai-nemo-think") 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 DigitalLearningGmbH/educa-ai-nemo-think with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "DigitalLearningGmbH/educa-ai-nemo-think" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "DigitalLearningGmbH/educa-ai-nemo-think", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/DigitalLearningGmbH/educa-ai-nemo-think
- SGLang
How to use DigitalLearningGmbH/educa-ai-nemo-think 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 "DigitalLearningGmbH/educa-ai-nemo-think" \ --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": "DigitalLearningGmbH/educa-ai-nemo-think", "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 "DigitalLearningGmbH/educa-ai-nemo-think" \ --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": "DigitalLearningGmbH/educa-ai-nemo-think", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use DigitalLearningGmbH/educa-ai-nemo-think with Docker Model Runner:
docker model run hf.co/DigitalLearningGmbH/educa-ai-nemo-think
educa-ai-nemo-think
This is a preliminary release of our internal testing reasoning model trained on a purely german dataset of reasoning traces mostly generated with DeepSeek R1.
Training details
We trained this model for 2 epochs on 10k reasoning SFT samples. No DPO / GRPO / other preference optimization has been done. The dataset is a POC subset of our much larger reasoning dataset which we are still filtering and cleaning. As such, this model will be updated soon when we have trained it on the full dataset.
Sampling Parameters
We've had success with top_p=0.95, top_k=30, repetition_penalty=1.15 and temperature=0.4 and it's also recommended to append <think> to force reasoning like this:
input_text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
input_text += "<think>\n"
input_ids = tokenizer(input_text, return_tensors="pt")["input_ids"].to("cuda")
Example outputs:
Human: Führe einen Induktionsbeweis und zeige, dass die Summe von i=1 bis n von 2i-1 = n^2 ist. Assistant: <think> Ich soll einen Induktionsbeweis führen, um zu zeigen, dass die Summe von 2i-1 für i=1 bis n gleich n^2 ist. Zuerst muss ich die Formel für die Summe von 2i-1 für i=1 bis n aufschreiben. Das ist die Summe von 1, 3, 5, ..., 2n-1. Ich denke, dass ich die Summe durch eine andere Methode auswerten kann, um zu zeigen, dass sie n^2 ergibt.
Eine Möglichkeit wäre, die Summe direkt auszurechnen, aber das wird mit zunehmender n sehr kompliziert. Eine andere Methode wäre, die Summe durch eine bekannte Formel auszudrücken. Ich erinnere mich, dass die Summe einer arithmetischen Reihe Summe = n/2 * (anfang + ende) ist. Hier ist der Anfang 1 und das Ende 2n-1. Aber ich muss sicherstellen, dass ich die Summe richtig berechne.
Lass mich die Summe mit der arithmetischen Reihe Formulierung berechnen:
Summe = n/2 * (1 + (2n-1)) = n/2 * (2n) = n^2
Das ergibt tatsächlich n^2. Also scheint die Aussage zu stimmen. Jetzt muss ich den Induktionsbeweis führen.
Ich werde in zwei Schritten vorgehen: Basisfall und Induktionsschritt.
Basisfall: n=1 Ich muss zeigen, dass die Summe für n=1 gleich 1^2 ist. Die Summe ist dann nur 2*1-1 = 1. Also ist die Summe für n=1 tatsächlich 1^2. Der Basisfall ist erfüllt.
Induktionsschritt: Ich muss annehmen, dass die Summe für n=k gleich k^2 ist und dann zeigen, dass die Summe für n=k+1 gleich (k+1)^2 ist.
Die Summe für n=k+1 ist die Summe von 1 bis 2k+1. Das ist die Summe von 1 bis 2k plus 2k+1. Nach der Induktionsannahme ist die Summe von 1 bis 2k gleich k^2. Also ist die Summe für n=k+1 gleich k^2 + 2k+1. Ich muss zeigen, dass das gleich (k+1)^2 ist.
(k+1)^2 = k^2 + 2k + 1
Das ist genau die Summe, die ich berechnet habe. Also ist die Summe für n=k+1 tatsächlich gleich (k+1)^2. Der Induktionsschritt ist erfüllt.
Da sowohl der Basisfall als auch der Induktionsschritt erfüllt sind, ist die Aussage für alle n bewiesen. </think> Um die Summe von (2i-1) für (i=1) bis (n) zu zeigen, dass sie (n^2) ergibt, führe ich einen Induktionsbeweis durch.
Basisfall (n=1): Die Summe ist dann (2\cdot1-1=1). Das ist gleich (1^2). Der Basisfall ist erfüllt.
Induktionsschritt: Angenommen, die Summe für (n=k) ist (k^2). Dann ist die Summe für (n=k+1) die Summe von (2i-1) für (i=1) bis (k+1). Das ist die Summe von (2i-1) für (i=1) bis (k) plus (2(k+1)-1=2k+1). Nach der Induktionsannahme ist die Summe für (i=1) bis (k) gleich (k^2). Also ist die Summe für (i=1) bis (k+1) gleich (k^2+2k+1). Das ist gleich ((k+1)^2). Der Induktionsschritt ist erfüllt.
Da sowohl der Basisfall als auch der Induktionsschritt erfüllt sind, gilt die Aussage für alle (n).
- Downloads last month
- 5
Model tree for DigitalLearningGmbH/educa-ai-nemo-think
Base model
mistralai/Mistral-Nemo-Base-2407