Instructions to use nicomedu-labs/NicomEdu-Alpha with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nicomedu-labs/NicomEdu-Alpha with Transformers:

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

pipe = pipeline("text-generation", model="nicomedu-labs/NicomEdu-Alpha", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("nicomedu-labs/NicomEdu-Alpha", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use nicomedu-labs/NicomEdu-Alpha with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/nicomedu-labs/NicomEdu-Alpha

SGLang

How to use nicomedu-labs/NicomEdu-Alpha 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 "nicomedu-labs/NicomEdu-Alpha" \
    --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": "nicomedu-labs/NicomEdu-Alpha",
		"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 "nicomedu-labs/NicomEdu-Alpha" \
        --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": "nicomedu-labs/NicomEdu-Alpha",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nicomedu-labs/NicomEdu-Alpha with Docker Model Runner:
```
docker model run hf.co/nicomedu-labs/NicomEdu-Alpha
```

NicomEdu-Alpha

NicomEdu-Alpha is an experimental general language model. It is specialized through fine-tuning for SS2 Mathematics classroom-style responses.

Alpha release status

This release is for experimentation and research. It is not a verified mathematics authority and is not approved for high-stakes educational decisions.

Known limitations

It can give incorrect mathematics answers.
It can repeat text or continue after a complete answer.
It can use inconsistent Markdown-style formatting.
It can answer non-mathematics prompts.
Its performance outside SS2 Mathematics was not evaluated.
Verify important answers independently.

Base model and adaptation

Base model: IFM/K2-Horizon-3.7B
Base revision: 633f52ad28b17edeabd82afc61d2d13b4c59a561
Adaptation: QLoRA merged into base weights
Training format: plain instruction, input, and output text
Training dataset: 229 private controlled records
Dataset SHA-256: 662dcba5475857d276f43400d1c26b1d988bb28bb344f9ab21badff3fd1865b6
Dataset composition: 141 teacher-assisted presentation rewrites and 88 unchanged mode-specific records
Excluded unresolved source records: 11

Evaluation summary

A fresh 12-prompt private evaluation found no K2 control tokens. It also found an incorrect brackets-equation answer, a non-responsive first-hint result, and repetition in some responses. These findings are why this is an Alpha release.

Loading

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "nicomedu-labs/NicomEdu-Alpha"

tokenizer = AutoTokenizer.from_pretrained(
    model_id,
    trust_remote_code=True,
)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

License and attribution

This release is distributed under Apache-2.0. It is derived from IFM/K2-Horizon-3.7B at the revision listed above. See LICENSE and NOTICE for terms and attribution.

Downloads last month: -

Safetensors

Model size

5B params

Tensor type

BF16