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

Libraries

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

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

pipe = pipeline("text-generation", model="domyn/Domyn-Edge-Teacher-IF", 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-IF", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

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

Use Docker

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

SGLang

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

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

Domyn Edge Teacher IF

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 instruction following teacher of Domyn Edge: Domyn Edge SFT after a specialized fine-tuning and GRPO round on instruction following and tool calling problems with verifiable final answers. 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 IF is a 4B-parameter dense transformer with the same architecture and tokenizer as Domyn Edge SFT. It was trained in two stages (SFT + GRPO) on top of Domyn Edge SFT:

SFT: ... TODO
GRPO: asynchronous group relative policy optimization on instruction following and tool calling problems with a single verifiable final answer.

The weights are stored in bfloat16.

How to use

The model loads and serves like Domyn Edge SFT; see the Domyn Edge SFT card for the recommended environment, Transformers and vLLM instructions. The one difference is the context length: this teacher's config sets max_position_embeddings to 131072, so serve it at 128k instead of 256k:

vllm serve domyn/Domyn-Edge-Teacher-IF \
  --trust-remote-code \
  --max-model-len 131072 \
  --reasoning-parser qwen3 \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder

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.

Output format. The model reasons first and ends with the final answer, following the format requested by the user.

Evaluation

Teacher vs student on the Domyn Edge evaluation suite

Scores of this teacher next to those of the student, Domyn Edge SFT. All scores are percentages. avg@10 is the accuracy averaged over 10 rollouts per problem.

Group	Benchmark	Metric	Domyn Edge SFT	Teacher IF	Teacher IF vs Domyn Edge SFT
Math	AIME 2026	avg@10	61.3	79.3	+18.0
Multilingual math	MGSM (de, en, es, fr, ru)	accuracy	85.8	90.4	+4.6
Knowledge and reasoning	GPQA Diamond	pass@1	56.3	64.3	+8.0
Instruction following	IFBench	prompt strict pass@1	24.1	66.0	+41.9
Function calling	BFCL v4, without web search (overall)	pass@1	36.5	43.1	+6.6
	BFCL v4 (non-live)	pass@1	51.9	86.9	+35.0
	BFCL v4 (live)	pass@1	75.9	80.3	+4.4
	BFCL v4 (multi-turn)	pass@1	35.9	41.1	+5.2
	BFCL v4 (memory)	pass@1	27.3	26.0	-1.3
Agentic	SWE-bench Verified	pass@1	33.6	40.2	+6.6
Code	LiveCodeBench	pass@1	73.0	76.2	+3.2

IF evaluation suite

All scores are percentages. avg@32 is the accuracy averaged over 32 samples per problem.

Benchmark	Metric	Domyn Edge SFT	Qwen3.5-4B	Teacher IF	Teacher IF vs Qwen3.5-4B
IFBench	prompt-level strict	24.1	52.7	66.0	+13.3
IFEval	prompt-level strict	75.3	85.2	92.8	+7.6
IFStruct	accuracy	52.7	70.6	92.2	+21.6
BFCL v4, without web search (overall)	accuracy	36.5	43.2	43.1	-0.1
BFCL v4 (non-live)	accuracy	51.9	75.8	86.9	+11.1
BFCL v4 (live)	accuracy	75.9	76.5	80.3	+3.8
BFCL v4 (multi-turn)	accuracy	35.9	45.4	41.1	-4.3
BFCL v4 (memory)	accuracy	27.3	31.0	26.0	-5.0

Evaluation settings

Served with vLLM in bfloat16 at 128k context, 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: 0.0
- repetition_penalty: 1.0
- frequency_penalty: 0.0
AIME and HMMT: 32 samples per problem, accuracy averaged over the samples.

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. Only instruction following and tool calling were trained beyond Domyn Edge SFT.

Long generations. On hard problems the model can reason for tens of thousands of tokens. Training capped responses at 64k tokens; cap the generation length.

Safety work is incomplete. No dedicated safety tuning or red teaming was done on this model.

Open resources

This teacher: domyn/Domyn-Edge-Teacher-IF
Its data: domyn/Domyn-Edge-Teacher-IF-Mix
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: 650

Safetensors

Model size

4B params

Tensor type

BF16

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

Base model

domyn/Domyn-Edge-Base

Finetuned

domyn/Domyn-Edge-SFT

Finetuned

(1)

this model

Collection including domyn/Domyn-Edge-Teacher-IF

Domyn Edge

Collection

17 items • Updated 1 day ago

domyn
/

Domyn-Edge-Teacher-IF