Instructions to use domyn/Domyn-Edge-Teacher-STEM with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use domyn/Domyn-Edge-Teacher-STEM with Transformers:

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

pipe = pipeline("text-generation", model="domyn/Domyn-Edge-Teacher-STEM", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("domyn/Domyn-Edge-Teacher-STEM", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use domyn/Domyn-Edge-Teacher-STEM with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/domyn/Domyn-Edge-Teacher-STEM

SGLang

How to use domyn/Domyn-Edge-Teacher-STEM 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 "domyn/Domyn-Edge-Teacher-STEM" \
    --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": "domyn/Domyn-Edge-Teacher-STEM",
		"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 "domyn/Domyn-Edge-Teacher-STEM" \
        --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": "domyn/Domyn-Edge-Teacher-STEM",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use domyn/Domyn-Edge-Teacher-STEM with Docker Model Runner:
```
docker model run hf.co/domyn/Domyn-Edge-Teacher-STEM
```

Domyn Edge Teacher STEM

Phased release. Domyn Edge is a rolling release in three bundles. This teacher is part of the second bundle, together with the other domain teachers and their post-training data. See Open resources for everything released so far.

This repository holds the STEM teacher of Domyn Edge: Domyn Edge SFT after a STEM-only supervised pass and GRPO on verifiable multiple-choice science questions. It is one of the domain teachers that score the student's rollouts token by token in the multi-teacher on-policy distillation (MOPD) stage of our post-training protocol. It is released for research and reproducibility; it is not a general-purpose assistant.

Model summary
How to use
Evaluation
Limitations
Open resources
License
Citation

Model summary

Domyn Edge Teacher STEM is a 4B-parameter dense transformer with the same architecture and tokenizer as Domyn Edge SFT. It was trained in two stages:

STEM SFT: one epoch at 64k context on a STEM-only mix (science, general reasoning QA and long-document QA).
GRPO: asynchronous group relative policy optimization on 48,329 multiple-choice science questions, with a binary reward on the extracted option.

The weights are stored in bfloat16.

How to use

The model loads and serves exactly like Domyn Edge SFT; see the Domyn Edge SFT card for the recommended environment, Transformers and vLLM instructions.

Chat template. The repository ships the Domyn Edge SFT chat template unchanged; it is the template used during GRPO training and in the evaluation below. It opens the system turn with a fixed identity line ("You are Domyn Edge, a model trained by Domyn.").

Reasoning length. On hard questions the model often reasons past 10k tokens, so leave room in max_new_tokens.

Evaluation

Teacher vs student on the general evaluation suite

Scores of this teacher next to those of the student, Domyn Edge SFT, with one benchmark per area, to check what the teacher keeps outside its training domain. All scores are percentages.

Group	Benchmark	Metric	Domyn Edge SFT	Teacher STEM	Δ
Math	AIME 2026	avg@10	61.3	61.3	+0.0
Multilingual math	MGSM (de, en, es, fr, ru)	accuracy	85.8	88.4	+2.6
Knowledge and reasoning	GPQA Diamond	pass@1	56.3	68.7	+12.4
Instruction following	IFBench	prompt strict pass@1	24.1	21.3	-2.8
Function calling	BFCL v4, without web search (overall)	pass@1	36.5	36.3	-0.2
	BFCL v4 (non-live)	pass@1	51.9	56.8	+4.9
	BFCL v4 (live)	pass@1	75.9	76.7	+0.8
	BFCL v4 (multi-turn)	pass@1	35.9	32.9	-3.0
	BFCL v4 (memory)	pass@1	27.3	27.3	+0.0
Agentic	SWE-bench Verified	pass@1	33.6	35.4	+1.8
Code	LiveCodeBench (full set)*	pass@1	73.0	71.0	-2.0

*LiveCodeBench (full set) is all 1,055 problems of LiveCodeBench v6, released between May 2023 and April 2025.

Evaluation settings

Served with vLLM in bfloat16, with the chat template shipped in this repository.
Thinking on. Sampling: temperature 1.0, top-p 0.95, top-k 20, min-p 0, no presence or repetition penalty.

STEM evaluation suite

Benchmark	Metric	Teacher STEM	Qwen3.5 4B	Δ
MATH	accuracy	93.0	96.7	-3.7
MGSM (de, en, es, fr, ru)	accuracy	88.4	90.2	-1.8
GPQA Diamond	pass@1	69.1	77.3	-8.2
MMLU-Pro STEM (avg)	accuracy	84.0	87.5	-3.5
MMLU-Pro STEM: math	accuracy	91.6	93.3	-1.7
MMLU-Pro STEM: biology	accuracy	86.5	89.5	-3.0
MMLU-Pro STEM: chemistry	accuracy	85.8	89.6	-3.8
MMLU-Pro STEM: physics	accuracy	87.4	89.7	-2.3
MMLU-Pro STEM: computer science	accuracy	80.0	81.2	-1.2
MMLU-Pro STEM: engineering	accuracy	66.9	75.4	-8.5
DROP	F1	83.1	86.3	-3.2
LiveBench math	pass@1	72.1	78.5	-6.4
LiveBench reasoning	pass@1	72.3	83.9	-11.6

Evaluation settings

Served with vLLM in bfloat16, with the chat template shipped in this repository.
Thinking on. Sampling: temperature 1.0, top-p 0.95, top-k 20, min-p 0, presence penalty 1.5.
Qwen3.5 4B is evaluated with the same settings, using its own chat template.

Limitations

A teacher, not an assistant. This model is one of the domain teachers of multi-teacher on-policy distillation (MOPD), the post-training stage in which the Domyn Edge student generates its own responses and each teacher scores them token by token in its own domain.

Safety work is incomplete. No dedicated safety tuning or red teaming was done on this model. Do not deploy it in a user-facing application without your own safety layer and evaluation.

Open resources

This teacher: domyn/Domyn-Edge-Teacher-STEM
Its data: domyn/Domyn-Edge-Teacher-STEM-Mix (SFT and RL splits)
Student SFT model: domyn/Domyn-Edge-SFT, and its mix domyn/Domyn-Edge-SFT-Mix
Collection: Domyn Edge on Hugging Face

License

This model is released under Apache 2.0.

Citation

@misc{domynedge2026,
  title  = {Domyn Edge},
  author = {Domyn},
  year   = {2026},
  note   = {Technical report, forthcoming in November 2026.},
}

Downloads last month: 548

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for domyn/Domyn-Edge-Teacher-STEM

Base model

domyn/Domyn-Edge-Base

Finetuned

domyn/Domyn-Edge-SFT

Finetuned

(6)

this model

Dataset used to train domyn/Domyn-Edge-Teacher-STEM

Collection including domyn/Domyn-Edge-Teacher-STEM

Domyn Edge

Collection

This the official Domyn Edge collection, check out the blog to know more about each releases bundle at • 17 items • Updated about 5 hours ago

domyn
/

Domyn-Edge-Teacher-STEM